IMPROVING THE ROBUSTNESS OF PERFUSION IMAGING WITH ARTERIAL SPIN LABELING MAGNETIC RESONANCE IMAGING HUAN TAN

Transcription

1 IMPROVING THE ROBUSTNESS OF PERFUSION IMAGING WITH ARTERIAL SPIN LABELING MAGNETIC RESONANCE IMAGING BY HUAN TAN A Dissertation Submitted to the Graduate Faculty of WAKE FOREST UNIVERSITY GRADUATE SCHOOL OF ARTS AND SCIENCES in Partial Fulfillment of the Requirements for the Degree of DOCTOR OF PHILOSOPHY Biomedical Engineering December 2010 Winston-Salem, North Carolina Approved By: Robert A. Kraft, Ph.D., Advisor, Chair James B. Daunais, Ph.D. Craig A. Hamilton, Ph.D. W. Scott Hoge, Ph.D. Robert J. Plemmons, Ph.D.

2 ACKNOWLEDGEMENTS I would like to thank my advisor, Dr. Robert Kraft for his guidance and friendship. I am grateful for his time and patience working with me over the course of my graduate career, and his continuous support and encouragement in my endeavors outside of the world of MRI. I cherish the countless hours spent together at the scanner solving problems. I have grown as an independent engineer and researcher, and I owe a great deal of that progression to Dr. Kraft s guidance and mentorship. I would like to thank Drs. James Daunais, Craig Hamilton, Scott Hoge, and Robert Plemmons for serving on my Ph.D. committee. I greatly appreciate the time and effort they have put in to improve the quality of my dissertation. I owe special thanks to Dr. Scott Hoge for his collaboration and expertise on parallel imaging that has helped me to overcome many research barriers. I would also like to give thanks to Dr. Craig Hamilton for insightful discussion and Dr. James Daunais for his continuous support on non-human primate studies. I have been fortunate to be a member of the best lab at Wake Forest University, Laboratory for Complex Brain Networks (LCBN). I cannot express in quantitative terms how much I have enjoyed working there. I must thank the director of LCBN, Dr. Paul Laurienti, for always being available to help and making such a relaxed and supportive research environment. I would like thank all current and previous members of LCBN, especially Dr. Jonathan Burdette, Dr. Satoru Hayasaka, Dr. Ann Peiffer, Debra Hege, and Ashley Morgan. ii

3 My accomplishments would not be possible without the love and support of my family and friends. To my mother and father, I thank you both for your unconditional love, support, encouragement and sacrifice. You have been the main motivation during my pursuit of this degree. To Donna and Robert, thank you for your tireless love and support of my decisions and everything you have done for me in the last eleven years. You are my family here in the states. For all my wonderful friends, thank you for making those years so special and I will cherish the memories from good times for the rest of my life. iii

4 TABLE OF CONTENTS ACKNOWLEDGEMENTS... ii LIST OF ABBREVIATIONS... vi LIST OF FIGURES... ix LIST OF TABLES... xi ABSTRACT... xii Chapter I. NON-GADOLINIUM (ARTERIAL SPIN LABELING) TECHNIQUE...1 Accepted in Functional Neuroradiology: Principles and Clinical Applications, June 2010 II. A FAST, EFFECTIVE F METHOD FOR IMPROVING CLINICAL PULSED ARTERIAL SPIN LABELING MRI...22 Published in Journal of Magnetic Resonance Imaging, April 2009 III. 3D GRASE PROPELLER: IMPROVED IMAGE ACQUISITION TECHNIQUE FOR ARTERIAL SPIN LABELING PERFUSION IMAGING...43 In Press in Magnetic Resonance in Medicine, December 2010 IV. ROBUST EPI NYQUIST GHOST ELIMINATION VIA SPATIAL AND TEMPORAL ENCODING (EPI-GESTE)...65 Published in Magnetic Resonance in Medicine, December 2010 V. PSEUDO-CONTINOUSL ARTERIAL SPIN LABELING...95 VI. PERFUSION PHANTOM VALIDATION USING ARTERIAL SPIN LABELING MRI iv

5 VII. CONCLUSION SCHOLASTIC VITA v

6 LIST OF ABBREVIATIONS ASL MRI CBF CBV MTT PLD ETL SNR PASL CASL VS-ASL RF FAIR PICORE EPISTAR Arterial Spin Labeling Magnetic Resonance Imaging Cerebral Blood Flow Cerebral Blood Volume Mean Transit Time Post Labeling Delay Echo Train Length Signal to Noise Ratio Pulsed Arterial Spin Labeling Continuous Arterial Spin Labeling Velocity Selective Arterial Spin Labeling Radio Frequency Flow-sensitive Alternating Inversion Recovery Proximal Inversion with Control for Off-Resonance Effects Echo Planar Imaging and Signal Targeting with Alternating Radio Frequency AHS FOCI QUIPSS Q2TIPS SAR MT Adiabatic Hyperbolic Secant Frequency Offset Correction Inversion QUantitative Imaging of Perfusion using a Single Subtraction QUIPSS II with Thin-slice TI1 Periodic Saturation Specific Absorption Rate Magnetic Transfer vi

7 ASSIST EPI TE TI PWI M 0 PACS μ σ GRASE Attenuating the Static Signal in Arterial Spin Tagging Echo Planar Imaging Echo Time Inversion Time Perfusion Weighted Image Proton Density Weighted Image Picture Archiving and Communication System Mean Standard Deviation Gradient Echo And Spin Echo PROPELLER Periodically Rotated Overlapping ParallEL Lines with Enhanced Reconstruction 3DGP FOV PLACE PAGE SENSE PSF GESTE SVD GSR DSC PET 3D GRASE PROPELLER Field Of View Phase Labeling for Additional Coordinate Encoding Phased Array Ghost Elimination Sensitivity Encoding Point Spread Function Ghost Elimination via Spatial and Temporal Encoding Singular Value Decomposition Ghost to Signal Ratio Dynamic Susceptibility Contrast Positron Emission Tomography vii

8 NHP PCASL VSS TOF ROI RPP MP-PCASL WM BOLD Non-Human Primate Pseudo-Continuous Arterial Spin Labeling Very Selective Suppression Time Of Flight Region Of Interests Relative Perfusion Proportion Multi-Phase Pseudo-Continuous Arterial Spin Labeling White Matter Blood Oxygenation Level Dependent viii

9 LIST OF FIGURES Page 1 Normal ASL CBF Map General Principle of ASL and Image Post-processing Perfusion Signal from a Single Control/Label Pair ASL Labeling Schemes PASL Filter Processing Steps Mean Coefficient of Variation for 200 Study Cases Clinical CBF Perfusion Cases Before and After ASL Filtering Image Quality Assessment Scores DGP Trajectory and Pulse Sequence DGP Reconstruction Process Diagram DGP and 3D GRASE Invivo CBF Maps Comparison Reformatted Views of 3DGP and 3D GRASE Invivo CBF Maps Diagram of EPI-GESTE Reconstruction Algorithm An Image Domain Comparison of the parallel MRI Calibration Estimated Line Shift Coefficient and Comparisons of GSR Image Reconstruction Comparison for Nyquist Ghost Correction Residual Artifact Comparison Perfusion Weighted Image Comparison Difference Image Comparison Phantom and Signal Variance Images Balanced PCASL Labeling Scheme and Pulse Diagram...99 ix

10 22 PCASL and PASL Images from Human Subject One PCASL and PASL Images from Human Subject Two PCASL and PASL Images from NHP Design and Actual Illustration of the Perfusion Phantom Locations of the Thermocouple on the Perfusion Tissue Video Screen Capture of the Red-Dye Experiment Estimated Relative Perfusion Proportion as a Function of Slice Thickness Simulated Inversion Response Curve as a Function of Phase Offset Illustration of MP-PCASL x

11 LIST OF TABLES Page 1 Q2TIPS-FAIR Implementation Parameters Image Quality and Artifact Assessments Significance Scores (P-Values) Mean CBF Measurements for All Subjects Perfusion Results from Thermodynamic Validation xi

12 ABSTRACT Tan, Huan IMPROVING THE ROBUSTNESS OF PERFUSION IMAGING WITH ARTERIAL SPIN LABELING MAGNETIC RESONANCE IMAGING Dissertation under the direction of Robert A. Kraft, Ph.D., Assistant Professor of Biomedical Engineering Tissue function depends heavily on perfusion, a process which brings the blood supply to the tissue through the arterial system and removes the metabolic by-products via the veins. Abnormalities and disruptions in this process can have profound effects. Cerebral blood flow (CBF), a quantitative perfusion measurement in the brain in units of milliliters of blood per 100 grams of tissue per minute, can be used to understand complex neurophysiology and reveal neuropathology. Cerebral perfusion is conventionally measured with the use of an exogenous contrast agent that is invasive and has limited repeatability. A new class of technique, known as arterial spin labeling (ASL), uses the water molecules in the arterial blood as an endogenous tracer to measure CBF via magnetic resonance imaging. ASL is therefore completely non-invasive and has been widely adapted in both clinical and research environments. The principle of ASL is based on acquiring a control image where the inflowing blood is fully relaxed, and a label image where the inflowing blood is inverted. When the label image is subtracted from the control image, the remaining signal is proportional to the blood that has perfused into the brain after the static tissue signal cancels out. The ASL technique has two components; the spin preparation that tags that arterial blood by xii

13 magnetic inversion or saturation, and the image acquisition that collects the imaging data. In this dissertation work, we have developed novel techniques to improve the robustness of ASL from both aspects. ASL technique was employed in the clinical environment at Wake Forest University Baptist Medical Center. With the initial experience of a large clinical population, ASL was found to be susceptible to hardware instability and patient motion. A post-processing filtering algorithm was developed to improve the overall perfusion image quality, including cases that were deemed uninterpretable. The filtering algorithm improved image quality in nearly 40% of the total population, and salvaged over 1000 previously unusable cases. In research, a novel image acquisition method, 3D GRASE PROPELLER (3DGP) was developed to address issues in ASL learned from the clinical experience. By combining a single-shot 3D acquisition method and a unique rotational trajectory, 3DGP improved the image resolution by 60% while reducing scan time by 50% compared to the clinical protocol. In addition to superior image quality, 3DGP was insensitive to patient motion and demonstrated higher perfusion sensitivity. Improvement was also made from the aspect of spin preparation. Pseudo-continuous ASL (PCASL) was a recently invented labeling technique that has shown optimal perfusion sensitivity. Not only PCASL has demonstrated improved perfusion sensitivity in human, it also provided a suitable labeling scheme for perfusion imaging in non-human primates (NHP). CBF was difficult to measure in NHP using conventional ASL method due to the small brain size. With PCASL, NHP perfusion images with good quality can be obtained without requiring special hardware. xiii

14 Lastly, the ASL technique development would benefit strongly from a phantom system that mimics the tissue perfusion process for validation and comparison. A flowbased perfusion phantom was built by researchers from Virginia Polytechnic Institute and State University and tested using a thermodynamic model. The phantom experiment using ASL showed no perfusion signal due to a long transit time as a result of limitations on current equipment and design imperfections. A new design and potential improvements of the phantom system is discussed in detail. xiv

15 CHAPTER I NON-GADOLINIUM (ARTERIAL SPIN LABELING) TECHNIQUE Huan Tan and Jonathan H. Burdette The following manuscript was accepted as a book chapter in Functional Neuroradiology: Principles and Clinical Applications (Springer). Stylistic variations are due to the requirement of the publisher. H. Tan prepared the manuscript and J.H. Burdette acted in an advisory and editorial capacity during manuscript preparation. Additional texts and figures are included by H. Tan to add clarifications in this dissertation. 1

16 INTRODUCTION After nearly two decades of development 1-4, arterial spin labeling (ASL) has become a widely adopted magnetic resonance imaging (MRI) method for quantitative measurement of cerebral blood flow (CBF). In comparison to the conventional techniques using intravascular contrast agents, ASL offers several advantages. First, ASL is completely non-invasive, requiring no injection of Gadolinium-based contrast agent. In addition to allowing repeatable scanning in human subjects, the lack of intravenous injection proves to be especially important for patients with chronic renal failure 5 and in the pediatric population where the use of radioactive tracers or exogenous contrast agents may be restricted. In this age of Nephrogenic Systemic Fibrosis (NSF), the non- Gadolinium based cerebral perfusion technique of ASL has proven to be quite successful without the worries associated with NSF in renal failure patients. Second, ASL offers absolute quantification of cerebral perfusion. It is more difficult to obtain accurate quantitative CBF values through bolus contrast tracking techniques, and, as such, most of these bolus contrast techniques are qualitative in the clinic, revealing relative changes in cerebral blood volume (CBV), CBF and mean transit time (MTT) 6-8. Absolute quantification allows for easy recognition of global perfusion abnormalities, as seen in several instances, such as hypercapnia or diffuse hypoxic/anoxic injury. Regional quantitative CBF assessments allow for comparisons between pre- and post-treatment states in patients undergoing chemotherapy, endarterectomy, thrombolysis, and therapy for migraines or seizures Other clinical applications demonstrating ASL usage include acute and chronic cerebrovascular diseases, neoplasms, epilepsy, and functional MRI 7, 13. The image quality and perfusion sensitivity of ASL methods have significantly 2

17 improved with recent technical advancements and hardware refinements (see Figure 1). Thus, increased ASL usage in the future is expected for both clinical and research applications. Figure 1. Normal ASL CBF map for a healthy person. METHODOLOGY ASL is based on a simple subtraction technique: Control images minus label images. During an ASL experiment, protons in the blood in vessels outside the imaging plane are labeled or unlabeled. Following a post-labeling delay (PLD) during which the labeled blood reaches the brain parenchyma; images are obtained of the parenchyma in a labeled and unlabeled or control state (see Figure 2). CBF maps can be obtained by scaling the control/label difference image appropriately. Varying the width and location 3

18 Figure 2. Illustration of general principle of ASL and image post-processing. of the labeling plane depends on the type of ASL technique used and will be discussed below. More specifically, the general principle of ASL is to use water molecules in the inflowing blood as an endogenous tracer to measure the blood that has entered the brain tissue. Perfusion weighting, known as labeling or tagging, is achieved by magnetically saturating or inverting MR signals of water protons in the arteries supplying 4

19 the tissue of interest. The region where the labeling process takes place is referred to as the tagging plane. Again, the width and location of the tagging plane vary depending on the specific ASL implementation. After waiting the short period of time or PLD, the tagged blood reaches and fully exchanges with the target tissue 14, where an image is then acquired. This image is known as the label image. In order to remove the signal from the static tissue in the label image, a control image is acquired where the inflowing blood is not tagged. By subtracting the label image from the control image, the remaining signal Δ M = M M is proportional to the local CBF. control label In a typical ASL experiment, the difference signal in a single subtraction pair is only a fraction (1 2%) of the corresponding tissue signal. Additional T1 dephasing of the tagged blood further reduces the measurable signal at the time of imaging (Figure 3). The fractional perfusion signal also depends on many parameters such as flow rate, T1 of the blood and tissue, and transit time for blood to travel from the tagging plane to the imaging region 4. In order to achieve adequate signal-to-noise ratio (SNR) for diagnostic purposes, multiple control/label pairs are acquired and signal averaged (e.g. 60 control/label pairs at Wake Forest University Baptist Medical Center where the imaging voxel size is 3.75 x 3.75 x 8 mm 3 for human clinical cases). The number of control/label pairs may increase or decrease to maintain appropriate SNR, depending on the voxel size. The exact MR imaging sequence for ASL consists of two distinct components, (1) a spin preparation component and (2) an image acquisition component. As the names imply, the inflowing blood is labeled into different magnetic states during spin preparation, while the actual imaging of the brain occurs during imaging acquisition after spin preparation. There are a variety of techniques for spin preparation and imaging 5

20 Figure 3. The difference signal between control and label images is proportional to CBF. Signal averaging is necessary to obtain adequate SNR for diagnostic purposes due to ASL s inherently low perfusion sensitivity. acquisition, each with its own merits and shortcomings, which will be discussed below. One important feature of the ASL sequence is that the two components are independent of each other, allowing researchers and clinicians to choose the best combination for their specific applications. Spin Preparation Schemes There are currently four main types of ASL spin preparation schemes: pulsed ASL (PASL), continuous ASL (CASL), pseudo-continuous ASL (PCASL) and velocityselective ASL (VS-ASL), all of which follow the general principles of ASL. They only differ in the ways in which the inflowing blood is magnetically tagged (Figure 4). The methodology, advantages and disadvantages of each scheme are discussed in the following paragraphs and are summarized in Table

21 Figure 4. The different labeling scheme for a variety of ASL preparation sequences. Pulsed ASL (PASL) saturates or inverts a slab of spins (stationary and moving) proximal to the imaging region. Continuous and pseudo-continuous ASL (CASL and PCASL) labels a narrow plane of spins continuously as opposed to PASL proximal to the imaging region. Velocity selective ASL (VS-ASL) selectively saturates flowing spins of certain velocity to measure perfusion. 1. PASL In PASL, short (5 to 20 milliseconds) radio frequency (RF) pulses are used to instantaneously invert a thick slab of spins (blood, tissue or both) to produce a difference in magnetization of the inflowing blood and the brain water 6. The tagging plane of PASL can be proximal to the imaging region or the entire brain. PASL methods such as FAIR 16 (flow sensitive alternating inversion recovery), PICORE 17 (proximal inversion with control for off-resonance effects) and EPISTAR 18 (echo planar imaging and signal targeting with alternating radio frequency) use a single inversion pulse during the spin preparation to magnetically label the flowing spins. The differences among these methods are mainly the location of the tagging plane and the magnetic state of the tagged spins for the control and label images. It is important to have a well-defined slice profile for the labeling RF pulses to eliminate residual signals from the static tissue. Those PASL methods were refined with the use of adiabatic hyperbolic secant (AHS) pulses 19, which are insensitive to B1 inhomogeneity, to improve the inversion efficiency and the slice profile. Further improvement in slice profile is achieved with FOCI (frequency offset 7

22 correction inversion) AHS pulses 19, 20. One limitation of PASL is the poorly defined distal edge of the tagging plane, which introduces systematic errors caused by spatially varying transit delay and flow-through effects that bias the perfusion quantification. This limitation is overcome with methods such as QUIPSS (quantitative imaging of perfusion using a single subtraction), QUIPSS II, and Q2TIPS (QUIPSS II with thin-slice TI1 periodic saturation) that apply saturation pulses between the labeling pulse and image acquisition to sharply define the distal edge of the tagging plane, thereby eliminating the systematic biases 21, 22. Overall, PASL techniques have high labeling efficiency and low RF power deposition compared to CASL or PCASL. However, the perfusion sensitivity is low due to increased transit delay time. 2. CASL CASL uses long and continuous RF pulses (1 2 seconds) along with a constant gradient field to induce an inversion in a narrow plane of spins, usually applied at the base of the brain 2, 23. The spins within a physiological range of velocities travelling through the tagging plane are inverted by adjusting the amplitude of the gradients and RF pulses through a phenomenon known as the flow-driven adiabatic inversion 6. Compared to PASL, higher perfusion sensitivity can be achieved with CASL due to its continuous inversion and closer position of the tagging plane with respect to the imaging plane. The primary drawback of CASL is the requirement of continuous RF transmitting hardware that is not commonly available on most commercial scanners. In addition, the long labeling pulses in CASL deposit a higher level of RF energy into the patient, which may exceed the FDA guidelines on specific absorption rate (SAR). Another effect of those 8

23 long RF pulses is to partially excite the imaging plane through magnetization transfer (MT) effects, for which special procedures and correction schemes are required 1, In general, the average labeling efficiency of CASL (80% - 95%) is lower than that of PASL (95%). Nonetheless, the tagging plane of CASL is closer to the imaging plane, minimizing the perfusion signal loss caused by T1 relaxation during the delivery of the blood, which compensates for the lower labeling efficiency. 3. PCASL PCASL was introduced to eliminate the CASL hardware requirement and reduce SAR without losing labeling efficiency. First developed by Garcia and colleagues 27, this technique uses a train of discrete RF pulses in combination with synchronous gradient fields to mimic the flow-driven adiabatic inversion seen in CASL, except that no special hardware is required. A standard body coil is used for transmission during spin preparation, and a phased array coil is used for reception. This technique provides a better balance between the tagging efficiency and the perfusion SNR, in addition to reduced magnetic transfer effects and lower RF power deposition. Depending on implementation, PCASL may be susceptible to B0 inhomogeneity and eddy currents VS-ASL Unlike the three previously discussed schemes, VS-ASL 29 selectively inverts spins based on blood velocity rather than spatial location. The blood with a higher velocity than a specific value is saturated by VS-ASL to achieve perfusion contrast. In theory, VS-ASL allows for a smaller and more uniform transit delay for the delivery of 9

24 the blood to the target tissue and provides CBF measurements under a slow and collateral flow condition, such as seen in stroke patients 29. It is difficult to determine the optimal value for the cutoff velocity. Inaccurate values chosen will lead to incorrect estimation of local perfusion and image artifacts. Saturation of signal from high velocity blood also results in lower SNR for VS-ASL than other methods. Apart from manipulating the magnetic states of the inflowing blood, other techniques have been investigated to improve perfusion sensitivity by suppressing the static tissue. Recall that ASL techniques are based on pair-wise subtraction of the control and label images acquired under alternating conditions with untagged and tagged inflowing blood. Under an ideal situation, the static tissue signal is canceled out and will not contribute to the perfusion signal. In reality, instabilities in the background signal and magnetic field fluctuation can add a substantial amount of noise to the difference signal, manifesting as subtraction artifacts and degrading the overall sensitivity of the ASL imaging. A recent approach 30 used multiple inversion background suppression pulses along with crusher gradients to minimize the destructive interference from the background tissue for 3D ASL imaging. This background suppression technique, termed ASSIST (Attenuating the Static Signal in Arterial Spin Tagging), has been demonstrated to improve the temporal stability of the perfusion series by nearly 50%. More recently, ASSIST has been modified and extended for multi-slice 2D ASL imaging 31. Imaging Acquisition Schemes Once the blood has been fully exchanged with the target tissue, the actual images are acquired during the imaging acquisition component. Due to the small amount of the 10

25 blood signal, it is desirable to use an imaging sequence with high SNR. The conventional acquisition method for ASL is 2D echo planar imaging (EPI) with the Cartesian trajectory for its fast acquisition speed and ease of reconstruction. The fast acquisition speed also reduces the potential for motion artifacts. Echo time (TE) plays a significant role in perfusion sensitivity because of T2-dependent signal decay. An optimal TE can be achieved using spiral trajectory to improve perfusion SNR. As a tradeoff for the fast acquisition speed, both EPI and spiral suffer from off resonance effects, resulting in geometrical distortion, regional blurring, and signal loss that significantly degrade the image quality. Note that in 2D multi-slice imaging, there exist multiple inversion times (TI) for different slices, which may result in signal decreases in more distal slices and difficulties in accurate perfusion quantification. To overcome such problems in 2D imaging, 3D imaging techniques have been used to boost perfusion SNR with fewer imaging artifacts. Namely, a single-shot 3D GRASE technique 32 has been demonstrated to provide a near 3-fold increase in perfusion SNR compared to EPI with the same imaging resolution and scan time. One advantage of 3D GRASE is the whole imaging volume is acquired with a single TI. All slices acquired with 3D GRASE hence have the same amount of perfusion weighting. 3D GRASE has also demonstrated less imaging distortion and fewer susceptibility artifacts compared with EPI. One disadvantage of 3D GRASE is increased through-plane blurring due to T2 decay. Parallel imaging, partial Fourier acquisition and alternative trajectory, such as PROPELLER 33 have been investigated to reduce T2 blurring and improve the overall acquisition efficiency of 3D GRASE. 11

26 CBF Quantification The difference (subtraction) signal from the control/label images is only a relative measure of the regional perfusion information. Buxton et al. 34 have developed a general kinetics model for the ASL signal combining the kinetics and relaxation process needed to extract the quantitative measurement of perfusion. An absolute quantitative perfusion map acquired with PASL can be obtained 21, 34 : CBF = ΔM ( TI2 ) 2 M αti q ( T, T, TI ) 0, blood 1 p 1, tissue 1, blood 2 e TI T 2 1, blood where CBF is the cerebral blood flow, ΔM(TI 2 ) (delta M) is the mean signal difference between of the control and label images, M 0,blood is the equilibrium magnetization of blood, α (alpha) is the tagging efficiency, TI 1 is the time duration of the tagging bolus, TI 2 is the inversion time of each slice, and T 1,blood is the longitudinal relaxation time of blood, and q p is a correction factor that accounts for the difference between the T 1 of blood and the T 1 of brain tissue. In our experiments, the T 1,blood is approximated from T 1,white matter, which is measured directly from a M 0 weighted image acquired with the perfusion images. All other parameters are known or assumed to be a constant (TI 1 =700ms, TI 1S =900ms, T 1, blood =1200ms). The quantitative CBF map measures perfusion in each voxel in units of milliliters of blood per 100 grams of tissue per minute. To improve the accuracy of CBF quantification, various studies have shown that a number of parameters must be taken into consideration, such as spin preparation scheme, arterial transit delay 14, 22, 35-37, tissue and blood properties, MT effects 1, 14, inversion efficiency 38-40, and capillary water permeability 41, 42. Understanding the influence and 12

27 the assumptions of the variables in the CBF quantification model yields more accurate interpretation of the final CBF values. Other ASL Considerations 1. Crusher Gradients Crusher gradients are commonly used in the ASL sequence to eliminate residual signals in the vasculature immediately before the image acquisition. Including the vascular signal will artificially increase the perfusion signal, thereby overestimating the true perfusion value. A benefit of the crusher gradients is to minimize quantification errors. However, due to the inherently low ASL signal, removing any signal from the image may degrade the overall image quality Post-Labeling Delay Time The post-labeling delay (PLD), the time allowing tagged blood to flow into and fully exchange with the tissue, is another important consideration for ASL studies. With a short PLD, the blood has not had sufficient amount of time to reach the tissue, resulting in low perfusion signal. While it may reduce the overall scan time, the final perfusion image with a short PLD may not reflect the accurate perfusion information, as most of the blood still remains in the vasculature, dephased by the crusher gradients. A long PLD allows more time for the inflowing blood to fully exchange with target tissue, leading to a more accurate global assessment of the cerebral perfusion. However, in such cases there is less available perfusion signal at the time of imaging due to longer time allowed for T1 13

28 relaxation. The optimal PLD depends on the age and health condition of the patients. In a clinical setting, the PLD should be determined according to the clinical population. 3. Higher Field Strength Magnets Perfusion sensitivity can benefit greatly from higher magnetic field strengths since the image SNR increases proportionally to the field strength. In addition, more perfusion signal is available at the time of imaging acquisition due to the prolonged T1. Not only does higher field strength improve the spatial and temporal resolution of the perfusion imaging, but longer PLD is also permitted to compensate for the delay transit effects commonly found in patients with strokes and other neurovascular diseases 8. The downside of higher magnetic field is the enhanced off-resonance effects, although ASL on 3T is becoming a standard practice with widespread usage. 4. Coils Phased-array coils are becoming increasingly available at most clinical centers, and these coils can be used to improve perfusion SNR. Furthermore, these phased-array coils can be used for parallel imaging to shorten effective echo train length (ETL) by the acceleration factor. Reduced ETL can optimize perfusion sensitivity via a shorter TE and less distortion as a result of off resonance effects. Although a reduction in SNR is usually associated with parallel imaging, in ASL imaging, some of the perfusion signal loss is recovered by the shortened TE, and the effects on CBF quantification is minimum

29 5. Post-Processing Filters Instabilities during scanning, such as patient motion and random system fluctuations, can cause errors in the control/label subtraction images. While only a small percentage of the control/label pairs are affected, those errors can sometimes manifest as large image artifacts, rendering the final CBF maps unusable. An example of an effective filtering technique 45 during post-processing is removing the corrupted data volumes by statistically analyzing the spatial and temporal information, thereby significantly improving the quality of the final perfusion image. In some cases, the ASL filtering technique has fully recovered CBF maps that were previously unusable. The implementation and validation of the ASL filter is discussed in detail in Chapter VI. CONCLUSIONS As ASL techniques become more widely available on clinical and research scanners, an understanding of the basic physics of the various techniques improves experimental design and clinical interpretation of the images. In this chapter, we reviewed the basics of ASL, including the various techniques for spin preparation and image acquisition. While each technique of ASL is slightly different, all techniques share the concept of subtraction of a control image from a labeled image. Due to the inherently small SNR in these subtracted images, this process of obtaining a perfusion image from subtraction must be repeated many times and with averaging of the resulting subtraction images to obtain an interpretable perfusion image. The non-invasive nature of this ASL process and the relatively easy quantification of the cerebral blood flow have made ASL a technique growing in popularity, and as ASL techniques continue to improve and 15

30 become more widely available on clinical MR scanners, ASL will undoubtedly become a standard clinical and research brain imaging sequence. 16

31 REFERENCES 1. Alsop DC, Detre JA. Multisection cerebral blood flow MR imaging with continuous arterial spin labeling. Radiology. Aug 1998;208(2): Detre JA, Leigh JS, Williams DS, Koretsky AP. Perfusion imaging. Magn Reson Med. Jan 1992;23(1): Williams DS, Detre JA, Leigh JS, Koretsky AP. Magnetic resonance imaging of perfusion using spin inversion of arterial water. Proc Natl Acad Sci U S A. Jan ;89(1): Calamante F, Thomas DL, Pell GS, Wiersma J, Turner R. Measuring cerebral blood flow using magnetic resonance imaging techniques. J Cereb Blood Flow Metab. Jul 1999;19(7): Sadowski EA, Bennett LK, Chan MR, et al. Nephrogenic systemic fibrosis: risk factors and incidence estimation. Radiology. Apr 2007;243(1): Liu TT, Brown GG. Measurement of cerebral perfusion with arterial spin labeling: Part 1. Methods. J Int Neuropsychol Soc. May 2007;13(3): Wintermark M, Sesay M, Barbier E, et al. Comparative overview of brain perfusion imaging techniques. J Neuroradiol. Dec 2005;32(5): Wolf RL, Detre JA. Clinical neuroimaging using arterial spin-labeled perfusion magnetic resonance imaging. Neurotherapeutics. Jul 2007;4(3): Pollock JM, Whitlow CT, Deibler AR, et al. Anoxic injury-associated cerebral hyperperfusion identified with arterial spin-labeled MR imaging. AJNR Am J Neuroradiol. Aug 2008;29(7): Pollock JM, Deibler AR, Burdette JH, et al. Migraine associated cerebral hyperperfusion with arterial spin-labeled MR imaging. AJNR Am J Neuroradiol. Sep 2008;29(8):

32 11. Deibler AR, Pollock JM, Kraft RA, Tan H, Burdette JH, Maldjian JA. Arterial Spin-Labeling in Routine Clinical Practice, Part 3: Hyperperfusion Patterns. AJNR Am J Neuroradiol. Mar Deibler AR, Pollock JM, Kraft RA, Tan H, Burdette JH, Maldjian JA. Arterial Spin-Labeling in Routine Clinical Practice, Part 2: Hypoperfusion Patterns. AJNR Am J Neuroradiol. Mar Latchaw RE, Yonas H, Hunter GJ, et al. Guidelines and recommendations for perfusion imaging in cerebral ischemia: A scientific statement for healthcare professionals by the writing group on perfusion imaging, from the Council on Cardiovascular Radiology of the American Heart Association. Stroke. Apr 2003;34(4): Alsop DC, Detre JA. Reduced transit-time sensitivity in noninvasive magnetic resonance imaging of human cerebral blood flow. J Cereb Blood Flow Metab. Nov 1996;16(6): Pollock JM, Tan H, Kraft RA, Whitlow CT, Burdette JH, Maldjian JA. Arterial spin-labeled MR perfusion imaging: clinical applications. Magn Reson Imaging Clin N Am. May 2009;17(2): Kim SG. Quantification of relative cerebral blood flow change by flow-sensitive alternating inversion recovery (FAIR) technique: application to functional mapping. Magn Reson Med. Sep 1995;34(3): Wong EC, Buxton RB, Frank LR. Implementation of quantitative perfusion imaging techniques for functional brain mapping using pulsed arterial spin labeling. NMR Biomed. Jun-Aug 1997;10(4-5): Edelman RR, Siewert B, Darby DG, et al. Qualitative mapping of cerebral blood flow and functional localization with echo-planar MR imaging and signal targeting with alternating radio frequency. Radiology. Aug 1994;192(2): Silver MS, Joseph RI, Hoult DI. Selective spin inversion in nuclear magnetic resonance and coherent optics through an exact solution of the Bloch-Riccati equation. Phys Rev A. Apr 1985;31(4):

33 20. Yongbi MN, Yang Y, Frank JA, Duyn JH. Multislice perfusion imaging in human brain using the C-FOCI inversion pulse: comparison with hyperbolic secant. Magn Reson Med. Dec 1999;42(6): Luh WM, Wong EC, Bandettini PA, Hyde JS. QUIPSS II with thin-slice TI1 periodic saturation: a method for improving accuracy of quantitative perfusion imaging using pulsed arterial spin labeling. Magn Reson Med. Jun 1999;41(6): Wong EC, Buxton RB, Frank LR. Quantitative imaging of perfusion using a single subtraction (QUIPSS and QUIPSS II). Magn Reson Med. May 1998;39(5): Detre JA, Alsop DC. Perfusion magnetic resonance imaging with continuous arterial spin labeling: methods and clinical applications in the central nervous system. Eur J Radiol. May 1999;30(2): Silva AC, Zhang W, Williams DS, Koretsky AP. Multi-slice MRI of rat brain perfusion during amphetamine stimulation using arterial spin labeling. Magn Reson Med. Feb 1995;33(2): Talagala SL, Ye FQ, Ledden PJ, Chesnick S. Whole-brain 3D perfusion MRI at 3.0 T using CASL with a separate labeling coil. Magn Reson Med. Jul 2004;52(1): Zhang W, Silva AC, Williams DS, Koretsky AP. NMR measurement of perfusion using arterial spin labeling without saturation of macromolecular spins. Magn Reson Med. Mar 1995;33(3): Garcia DM, Bazelaire CD, Alsop D. Pseudo-continuous Flow Driven Adiabatic Inversion for Aterial Spin Labeling: ISMRM 05; Wu WC, Fernandez-Seara M, Detre JA, Wehrli FW, Wang J. A theoretical and experimental investigation of the tagging efficiency of pseudocontinuous arterial spin labeling. Magn Reson Med. Nov 2007;58(5): Wong EC, Cronin M, Wu WC, Inglis B, Frank LR, Liu TT. Velocity-selective arterial spin labeling. Magn Reson Med. Jun 2006;55(6):

34 30. Ye FQ, Frank JA, Weinberger DR, McLaughlin AC. Noise reduction in 3D perfusion imaging by attenuating the static signal in arterial spin tagging (ASSIST). Magn Reson Med. Jul 2000;44(1): St Lawrence KS, Frank JA, Bandettini PA, Ye FQ. Noise reduction in multi-slice arterial spin tagging imaging. Magn Reson Med. Mar 2005;53(3): Gunther M, Oshio K, Feinberg DA. Single-shot 3D imaging techniques improve arterial spin labeling perfusion measurements. Magn Reson Med. Aug 2005;54(2): Pipe JG. Motion correction with PROPELLER MRI: application to head motion and free-breathing cardiac imaging. Magn Reson Med. Nov 1999;42(5): Buxton RB, Frank LR, Wong EC, Siewert B, Warach S, Edelman RR. A general kinetic model for quantitative perfusion imaging with arterial spin labeling. Magn Reson Med. Sep 1998;40(3): Gonzalez-At JB, Alsop DC, Detre JA. Cerebral perfusion and arterial transit time changes during task activation determined with continuous arterial spin labeling. Magn Reson Med. May 2000;43(5): Yang Y, Engelien W, Xu S, Gu H, Silbersweig DA, Stern E. Transit time, trailing time, and cerebral blood flow during brain activation: measurement using multislice, pulsed spin-labeling perfusion imaging. Magn Reson Med. Nov 2000;44(5): Ye FQ, Mattay VS, Jezzard P, Frank JA, Weinberger DR, McLaughlin AC. Correction for vascular artifacts in cerebral blood flow values measured by using arterial spin tagging techniques. Magn Reson Med. Feb 1997;37(2): Maccotta L, Detre JA, Alsop DC. The efficiency of adiabatic inversion for perfusion imaging by arterial spin labeling. NMR Biomed. Jun-Aug 1997;10(4-5): Roberts DA, Detre JA, Bolinger L, Insko EK, Leigh JS, Jr. Quantitative magnetic resonance imaging of human brain perfusion at 1.5 T using steady-state inversion of arterial water. Proc Natl Acad Sci U S A. Jan ;91(1):

35 40. Utting JF, Thomas DL, Gadian DG, Ordidge RJ. Velocity-driven adiabatic fast passage for arterial spin labeling: results from a computer model. Magn Reson Med. Feb 2003;49(2): Ewing JR, Cao Y, Fenstermacher J. Single-coil arterial spin-tagging for estimating cerebral blood flow as viewed from the capillary: relative contributions of intra- and extravascular signal. Magn Reson Med. Sep 2001;46(3): Parkes LM, Tofts PS. Improved accuracy of human cerebral blood perfusion measurements using arterial spin labeling: accounting for capillary water permeability. Magn Reson Med. Jul 2002;48(1): Deibler AR, Pollock JM, Kraft RA, Tan H, Burdette JH, Maldjian JA. Arterial Spin-Labeling in Routine Clinical Practice, Part 1: Technique and Artifacts. AJNR Am J Neuroradiol. Mar Fernandez-Seara MA, Edlow BL, Hoang A, Wang J, Feinberg DA, Detre JA. Minimizing acquisition time of arterial spin labeling at 3T. Magn Reson Med. Jun 2008;59(6): Tan H, Maldjian JA, Pollock JM, et al. A fast, effective filtering method for improving clinical pulsed arterial spin labeling MRI. J Magn Reson Imaging. May 2009;29(5):

36 CHAPTER II A FAST, EFFECTIVE FILTERING METHOD FOR IMPROVING CLINICAL PULSED ARTERIAL SPIN LABELING MRI Huan Tan, Joseph A. Maldjian, Jeffrey M. Pollock, Jonathan H. Burdette, Lucie Y. Yang, Andrew R. Deibler, Robert A. Kraft The following manuscript was published in Journal of Magnetic Resonance Imaging, volume 29 issue 5, pages Stylistic variations are due to the requirement of the publisher. H. Tan developed the technique, conducted all the experiments and analysis, and prepared the manuscript. The method evaluation was performed by Dr. J.A. Maldjian, Dr. J.M. Pollock, Dr. J.H. Burdette and Dr. L.Y. Yang. Dr. R.A. Kraft acted in an advisory and editorial capacity during manuscript preparation. 22

37 ABSTRACT Purpose: To evaluate the effectiveness of a fully automated post-processing filter algorithm to improve the robustness and reduce imaging artifacts in pulsed arterial spin labeling (PASL) MRI perfusion images in a large clinical population. Materials and Methods: A mean and standard deviation based filter was implemented to remove outliers in the set of perfusion weighted images (control label) before being averaged and scaled to quantitative cerebral blood flow (CBF) maps. Filtered and unfiltered CBF maps from 200 randomly selected clinical cases were assessed by four blinded raters to evaluate the effectiveness of the filter. Results: The filter salvaged many studies deemed uninterpretable secondary to motion artifacts, transient gradient and/or RF instabilities, and unexpected disruption of data acquisition by the technologist to communicate with the patient.the filtered CBF maps contained significantly (p < 0.05) less artifacts and were more interpretable than unfiltered CBF maps as determined by one tail paired t-test. Conclusion: Variations in MR perfusion signal related to patient motion, system instability, or disruption of the steady state can introduce artifacts in the CBF maps which can be significantly reduced by post-processing filtering. Diagnostic quality of the clinical perfusion images can be improved by performing selectively averaging without a significant loss in perfusion SNR. Key Words: Clinical Perfusion MRI, Arterial Spin Labeling, PASL Filtering 23

38 INTRODUCTION Arterial spin labeling (ASL) is a non-invasive magnetic resonance imaging (MRI) method for measuring cerebral blood flow (CBF). Recent studies 1 have shown great potential for ASL in clinical imaging. ASL is completely non-invasive, repeatable, quantitative, and has shown close correlations with other blood flow imaging techniques including Dynamic Susceptibility Contrast MRI, Computed Tomography perfusion, Positron Emission Tomography and Single Photon Emission Computed Tomography 2. While other imaging modalities use an exogenous contrast agent, ASL uses the water in the blood as an endogenous tracer to measure blood flow from a set of tagged (label) and untagged (control) images. The control and label image pairs are subtracted from each other to eliminate the static tissue signal while retaining the signal from the blood that has perfused into the imaging slices. Since the static tissue signal is much larger than the signal from the blood, instabilities between the control/label images resulting from patient motion, physiological conditions, hardware instabilities, and unexpected mid-scan interruptions by the technologist to communicate with the patient can lead to large subtraction errors. While these instabilities may occur in only a small percentage of the control / label images, the effect can have a drastic and detrimental affect on the final perfusion weighted image (PWI) obtained by averaging all of the individual PWIs together in the time series. These subtraction errors between the control and label images can manifest themselves in the individual PWIs as hyper- or hypointensive perfusion signal involving just a few slices or the entire brain volume. By selectively discarding these inconsistent images in the perfusion time series, the final 24

39 averaged PWI can be improved with only a minor penalty in the signal to noise ratio (SNR). A number of techniques have been developed to improve CBF maps acquired with ASL: physiological noise reduction for functional perfusion experiments 3, 4, applying a low pass filter to the subtracted control / label images 5, static tissue background suppression methods 6, and selective averaging of control / label images based on motion criteria 7. While these methods have been shown to be of importance when examining the spectrum of the perfusion time series for functional MRI ASL studies, they may not be able to correct for dramatic signal variations caused by hardware instability, large and abrupt patient motion, and unexpected disruption of the steady state. One previous study 7 has shown that the quality of CBF maps can be improved by removing control / label pairs that were acquired during periods of motion exceeding 2mm of translations and 1.5 degrees of rotations. The remaining control / label images were subtracted from each other and averaged to create a final PWI. The final PWIs for all cases were visually inspected. Those with severe residual artifacts were removed from the study for further analysis. For research studies with a dedicated staff team, visual inspection is certainly a viable method for post-processing and improving perfusion image quality. However, as perfusion imaging becomes readily available for clinical applications, it is neither feasible nor practical to manually process each ASL case. An automated method is necessary for improving the perfusion image quality in a clinical environment. At our institution, a pulsed ASL (PASL) CBF map is acquired as part of the standard clinical MRI protocol for patients with an indication of vascular abnormalities 25

40 (stroke, TIA, or carotid stenosis), neoplasm or seizure. Over clinical CBF maps 8-10 with a rate of cases per day have been acquired in a time period of two years using an automated image processing pipeline 11. This represents the largest experience with PASL in clinical practice to date. In this paper, we describe an automated postprocessing filtering procedure that effectively improves the overall quality and stability of the clinical perfusion measurement. MATERIALS AND METHODS PASL Acquisition and Reconstruction After obtaining IRB approval from our institution, this study used data from our clinical patient population undergoing an MRI examination. The purpose was to evaluate the effectiveness of a post-processing filter to improve perfusion images acquired for diagnostic purposes. Image acquisitions were done among five clinical MRI scanners (three 1.5T GE Signa Software Platform 12M5, one 1.5T GE Twinspeed Software Platform 14M5, and one 3T GE Twinspeed Software Platform 12M5, GE Healthcare, Milwaukee, WI). Body coil transmit with receive-only head coils (HD 8 Channel HiRes Brain Array Coil and 4 Channel Neurovascular Coil manufactured by Invivo Devices, Gainesville, FL) were used for data collection. Cerebral blood flow (CBF) was measured with QUantitative Imaging of Perfusion using a Single Subtraction with Thin Slice TI1 Periodic Saturation (QUIPSS II TIPS a.k.a. Q2TIPS) 12, 13 with Flow-sensitive Alternating Inversion Recovery (FAIR) 14 acquiring 60 control / label pairs in 6.3 minutes using a single shot, ramp sampled gradient echo EPI. The pulse sequence was developed and validated internally. The acquisition parameters 26

41 are listed in Table 1. A proton density weighted (M 0 ) image was acquired at the beginning of the Q2TIPS-FAIR sequence for scaling the final PWI to a quantitative CBF map. Table 1. Q2TIPS-FAIR implementation parameters Parameter Value FOV / Slice Thickness 240mm / 8mm TI 1 / TI 1s / TI 2 / TR 800ms / 1200ms / 2000ms / 3000ms Acquisition Matrix 64 x 64 TE 24.1 ms Number of Slices Varying from 7-17 b (diffusion gradient) 5.25 mm 2 /sec The imaging data was automatically transferred from the MRI scanners to the offline workstations for fully automated reconstruction, using the Sun Grid Engine (Sun Microsystems, Santa Clara, CA) for distributed processing. The individual control and label imaging volumes were realigned with the software package SPM2 (Statistical Parametric Mapping, Wellcome Department of Imaging Neuroscience, London, UK). The PWI was then scaled to a quantitative CBF map using the General Kinetic Model as described by Buxton et al 15. Magnetic field strength differences were accounted for during quantification by assuming different relaxation times for the blood (T 1 of the blood was assumed to be 1200 ms at 1.5T and 1490 ms at 3T 16 ). As part of the automated processing pipeline, the final CBF maps were transferred to the clinical Picture Archiving and Communications System (PACS, AGFA, Mortsel, Belgium) for system wide review. 27

42 ASL Filtering In PASL, the overall MR signal by the delivered blood is about 1% of the total signal of the tissue, resulting in very low SNR for a single subtraction image 17. The control / label pairs are thus acquired multiple times such that the subtraction images are averaged together to achieve proper SNR for diagnostic purposes. Since the control / label images are acquired under identical conditions, the resulting PWIs should be identical with the exception of noise. However, we have observed that a few PWIs can have significant signal variation leading to a variety of imaging artifacts, and thereby rendering the CBF results of poor diagnostic quality. The quality of the CBF map can be significantly improved by selectively removing those corrupted PWIs. Our ASL filtering technique was designed to improve the quality of the final perfusion image by automatically identifying and discarding the inconsistent PWIs from the control / label subtractions. Before filtering, a brain tissue mask was created by averaging all the control and label images together. Pixels of values greater than 5% of the maximum of the averaged image were classified as brain tissue, whereas the remaining pixels were classified as background. The tissue mask was dilated with a 4-pixel wide kernel to ensure the entire brain coverage. Two criteria, the mean and the standard deviation, were implemented in the filter. The mean criterion detected PWIs with artificial hyper- or hypo-intensity. The mean intensity of voxels classified as tissue from the PWI, denoted asδ Mi, was calculated. The mean ( μ Δ M ) and the standard deviation ( σ Δ M ) of Δ Mi for all the PWIs were calculated. Individual PWIs with mean tissue value Δ Mi that satisfied 28

43 Δ M > μ + n σ [1] i ΔM ΔM were classified as outliers and were removed from the set of PWIs to be averaged together. The parameter n represents a scaler. A threshold of n = 2.5 was chosen based on empirical performances of the filter from over 100 initial clinical cases prior to the statistical evaluation (see RESULTS). In the ideal situation, the signal variation among tissue voxels in a single PWI is expected to be similar to the background noise due to the low SNR. The standard deviation criterion was designed to identify and remove PWIs with large signal variation in the region classified as tissue when compared with the average PWIs. The standard deviation of the tissue voxels in the PWI, denoted as S i, was calculated. The mean ( μ S ) and the standard deviation ( σ S ) of S i across the set of PWIs were calculated. Individual PWIs with standard deviation value S i that satisfied S > μ + m σ [2] i S S were classified as outliers and removed from the set of PWIs to be averaged together. The parameter m represents a scaler and m= 1.5 was determined based on empirical performances of the filter from the same initial dataset as the mean criterion that was used to determine the scaler variable n. To minimize over filtering, a PWI was considered stable when the natural log of the difference value between max( S i ) and min( S i ) was less than 1, that is, ln(max( S ) min( S )) < 1 [3] i i 29

44 Figure 5. PASL filter processing steps. The tissue data is extracted by masking out the background noise. The filtering based on the mean and standard deviation is performed in both volume-based and slice-based fashions, for which the union of the results determines the subtraction images that are discarded from the final calculation of the perfusion-weighted image. When this criterion was satisfied, all of the individual PWIs were included in the final averaging step. The software for the ASL filter was written in Matlab (MathWorks, Natick, MA) and inserted into our automated PASL post-processing pipeline. The overall design of the ASL filtering process is illustrated in Figure 5. Two criteria, the mean and the standard deviation across the set of PWIs, were executed in parallel to remove inconsistent PWIs as previously described. It was also observed that 30

45 instability could affect all or only a few slices within the brain volume. Hence, the filtering procedure was performed on both a volume-by-volume basis and a slice-by-slice basis. This allowed removal of an entire volume, or just a few slices within a single volume of the corrupted dataset. In the volume-based approach, all the slices were treated together as a single volume for which Δ Mi and S i were calculated along the temporal domain. In the slice-based approach, Δ Mi and S i were calculated for each individual slice along the temporal domain. The remaining images that were not identified as outliers were used for the calculation of the final averaged PWI. The SNR of the final averaged PWI undergoing filtering is slightly reduced according to SNR Loss 1 N N filtered total where N total is the total number subtraction images before filtering and N filtered is the actual number of volumes used for averaging after filtering. On average the ASL filter removed 5 images per slice, corresponding to an SNR loss of 4%. Image Quality Assessment and Statistical Analysis CBF maps of 200 clinical cases randomly selected from a total of 540 cases performed in February, 2008, were to be included in the study. The patient population of those 200 cases was representative of the clinical patients who had CBF maps acquired as part of their MRI examinations. Briefly, the study included 88 males and 112 females ranging from 0.3 to 91 years old with a mean age of 48. From these 200 cases, the MRI examinations as determined by a board certified radiologist revealed that 33% were considered normal (i.e. no clinical findings); 14.5% had tumor; 11% had undergone brain surgical procedures; the remaining cases consisted of pathologies including seizure, 31

46 infarct, atrophy, metastases, hemorrhage, arachnoid cyst, arterio-venous malformation, and anoxia. The filtered and unfiltered CBF maps from these patients were independently analyzed in a random order by three board-certified neuroradiologists experienced in clinical PASL interpretation and one board certified physician with significant radiology training and experience with research PASL analyses. The observers were blinded to the medical history of the patients and evaluated the CBF images on image quality criteria only. Intraobserver variability was measured by having 50 cases randomly selected out of the 200 to be rated twice. Prior to the individual analysis, all observers had a training session discussing the baseline rules for quality assessment followed by a trial dataset for quality assurance. The observers were asked to rate the individual images based on the overall image interpretability and three representative artifacts including edge, ring and CSF shine through (see Figure 3, details of the artifacts were previously described 8 ). Adopted from McConnell et al 18, each potential artifact was rated on a scale of 0 3 where 0 indicated no gross artifacts and 3 indicated severe artifacts, except for CBF shine through which was either present or absent. To maintain consistency with artifact scores for the overall evaluation, we decided to use 0 for the most interpretable case and 3 for the least interpretable case. Rating details are described in Table 2. Table 2. Image Quality and Artifact Assessments Interpretability Edge & Ring CSF Shine Through Scale Grading Scale Grading Scale Grading 0 Interpretable 0 None 0 Non-exist 1 Interpretable with minor artifacts 1 Minor 1 Exist 2 Interpretable with major artifacts 2 Moderate 3 Uninterpretable 3 Severe 32

47 The image quality scores in each category for the filtered and unfiltered CBF maps were calculated by averaging rating scores from all raters. A lower score indicated a more interpretable image with fewer artifacts. Since the primary function for the ASL filter was to remove outliers in the perfusion time series based on certain characteristics, the resulting time series would have less variation and hence, would be more stable. We wanted to focus on detecting improvements that the ASL filter has provided. To do so, a one-tail, paired t-test was performed to compare the ratings on the filtered and unfiltered CBF maps. p < 0.05 was considered significant for filtered CBF maps being easier to interpret with fewer artifacts. A two sample t-test assuming equal variance was used to examine the intraobserver variability on the repeated 50 cases. If p > 0.05, then there was no significant difference between the rating scores from the first and second repeated trial for the same observer. The intraobserver statistical tests were two-tailed. The rating process was done via internally developed software. 33

48 Figure 6. Mean coefficient of variation (CV) values were calculated for gray matter voxels for each perfusion case. Standard deviations (SD images) were calculated across perfusion time series to visually represent cases with large CV (magnitude) values. All SD images were normalized for comparison purposes. Three sample cases from which the ASL filter had no influence in the final CBF, completely recovered the perfusion signal, and recovered partial perfusion signal were shown in sub figure A, B and C respectively. It should be noted that CV can be used as one, but not an exclusive, factor to quantify the stability of a perfusion case. Although there was a trend that cases with high (magnitude) CV were less stable, exceptions may be made for patient with hypo-perfusion. CV values were calculated for 185 cases out of 200. The remaining cases were excluded because the gray matter mask was either unavailable or visually inaccurate. RESULTS Filtering Results As previously discussed, signal across PWIs should contain little variation for a stable case under ideal situation. Any unexpected fluctuation in the system could cause drastic signal changes in the perfusion time series producing various artifacts. Those fluctuations were seen by computing the standard deviation of the perfusion signal across all time points. The coefficients of variation of gray matter voxels were calculated along temporal domain and averaged for all the cases to demonstrate how system fluctuations 34

49 could affect the CBF maps (Figure 6). The causes for those system fluctuations were speculated to be primarily associated with patient motion, hardware instability, or a disruption of the steady state by the technologist to communicate with the patient. The ASL filter has removed those artifacts in a significant number of cases. Examples of unstable and stable clinical perfusion cases before and after filtering are illustrated in Figure 7. Many of the unstable cases before filtering presented with little usable information. After removing a few corrupted volumes (> 95% of the original data was kept), the perfusion information was recovered without significant loss of SNR. Unstable perfusion cases for which the ASL filter was ineffective were usually associated with continuous large movements from the patient throughout the data acquisition. Artificial hyper- and hypo-perfusions could be observed among most PWIs in the perfusion time series leaving little usable information and those cases were usually unrecoverable. Such scenarios usually required rescan. It should be noted that not all hypo-perfusion cases were artificial. Some may be associated with true physiological conditions that would be inherently difficult to measure with ASL methods. In many of these cases, the low ASL signal in the unfiltered and filtered maps was the reason why the filtered CBF maps were rated lower than the unfiltered CBF maps. 35

50 Figure 7. Examples of clinical CBF maps before and after ASL filtering for both unstable and stable cases. The scale for the color map is in milliliters of blood / 100 grams of tissue / minute. ASL Filter Evaluation Results Excellent agreement was found within the repeated cases for all four observers (p > 0.5). The image quality scores for filtered CBF maps were better than unfiltered CBF maps across all categories. The individual scores for each observer were plotted in Figure 8, and it illustrates that the filtered CBF maps had consistently better scores than the unfiltered CBF maps, although there existed some variations in the baseline scores among different observers. Overall, four raters (22%, 52%, 49% and 26%) determined 36

51 Figure 8. Bar charts showing image quality scores based on each of the four categories, interpretability, edge artifact, ring artifact and CSF shine through artifact, for all observers. * - The rating score is significant that the filtered CBF map is better than unfiltered CBF map. - The rating score indicates there is a trend that the filtered CBF map is better than unfiltered CBF map. that on average 37% of all CBF maps (filtered and unfiltered) contained image artifacts. That is to say that a rater gave a score greater than 0 (no artifacts) for at least one of the three artifacts (edge, ring and CSF shine through) being assessed. On average among those images affected by artifacts, 39% were improved by the filtering process (40%, 37%, 41% and 38%), 45% were unaffected by the filtering process (46%, 47%, 42% and 46%). The remaining 16% of the cases where artifacts appeared to be worse with filtering were usually associated hypoperfusion and severe artifacts (the interpretability scores for those cases were rated equal or greater than 2) and belonged to the unrecoverable category. 37

52 According to the one-tail paired t-test, the ASL filter significantly reduced the ring artifact (p < 0.05) and completely eliminated CSF Shine Through (p = 0). Three observers found the ASL filter significantly reduced edge artifacts (p < 0.03) while one observer found there was a tendency for the ASL filter to reduce edge artifact (p ~ 0.06). Three observers found the ASL filter significantly improved the interpretability of the CBF maps (p < 0.04) while one neuroradiologist disagreed on the same issue (p > 0.1). The p values for each rater were shown in Table 3. We have observed that when the SNR of the CBF maps were low, some observers preferred unfiltered images with slightly higher SNR compared to the filtered image when artifacts were minor. This may explain the fluctuation in the results for image interpretability. Table 3. Significance scores (p-values) of each rater for all categories Observer Interpretability Edge Ring CSF Shine Through One Two Three Four DISCUSSION The purpose of this work was to evaluate a method to improve the quality and stability of the absolute quantitative perfusion maps in a clinical environment where PASL is used routinely. The conventional PASL sequence was used in conjunction with an additional post-processing filtering. The key design for the filter was to automatically detect inconsistency within the PASL time series and to improve the quality and robustness of the CBF map by removing corrupted data. A primary source of image artifacts in our CBF maps was related to patient motion that led to incomplete removal of the static tissue signal by imperfect subtraction 38

53 of the control / label pairs. Background suppression methods such as ASSIST 6 have been introduced to help reduce these subtraction errors by applying additional inversion pulses to null out static tissue. Even with background suppression, a small amount of the static tissue signal (~5%) is preserved in order to correct for patient motion. The residual static tissue signal after subtracting the control / label images, while significantly reduced, may still introduce some errors into the final perfusion image. Since the proposed filter in this paper is completely done after image acquisition, we anticipate that this filter may further improve the diagnostic image quality of perfusion images acquired with background suppression techniques. Our clinical experience with perfusion imaging has demonstrated that hardware system instability and disruption of the steady state occur in a small percentage of our clinical cases. For example, in Figure 3 image panel 5, a severe line artifact can be seen across the image. This was the result of the gradient fault of the phase encode gradient amplifier. In this particular case, the scanner continued to acquire data. While background suppression would reduce the signal from the static tissue, it would not remove an unforeseen artifact caused by RF / gradient instabilities or technologists pausing and resuming the PASL acquisition due to uncooperative patients. While these instances do not occur often, they happen frequently enough to warrant the inclusion in the proposed filtering method to salvage these images. In conclusion, we have developed and implemented a post-processing filtering technique to optimize clinical perfusion imaging. The ASL filter was shown to significantly reduce motion and system related image artifacts. Mechanisms were implemented to preserve good data and effectively prevent overfiltering. In stable 39

54 perfusion cases the filter had little effect on the final CBF map. The filter processing time per case was on the order of seconds, adding no additional computational load to the processing pipeline. The implementation of the ASL filter was purely in the postprocessing stream, meaning the filter design could be incorporated into any ASL scheme without being restricted to a particular preparation and acquisition technique. In conclusion, the ASL filtering technique has proven to be very clinically successful, salvaging perfusion images that would have been uninterpretable or difficult to interpret and making them clinically interpretable. In we have conducted over 10,000 clinical PASL scans. The ASL filter has become a necessity in improving the diagnostic quality and reducing image artifacts. The ASL filter has salvaged approximately 4% of the total cases in which CBF maps were completely uninterpretable to be diagnostically relevant. The ASL filter described here obviates the need to rescan patients in the event of an initially unusable scan, thus, implementation of such a filter can save time and resources for the patient and the hospital. ACKNOWLEDGEMENTS We would like to thank Kathy Pearson and Ben Wagner for the help with computer programming and Dr. Hayasaka for helping with the statistical analysis. 40

55 REFERENCES 1. Brown GG, Clark C, Liu TT. Measurement of cerebral perfusion with arterial spin labeling: Part 2. Applications. J Int Neuropsychol Soc. May 2007;13(3): Wintermark M, Sesay M, Barbier E, et al. Comparative overview of brain perfusion imaging techniques. J Neuroradiol. Dec 2005;32(5): Restom K, Behzadi Y, Liu TT. Physiological noise reduction for arterial spin labeling functional MRI. Neuroimage. Jul ;31(3): Pfeuffer J, Van de Moortele PF, Ugurbil K, Hu X, Glover GH. Correction of physiologically induced global off-resonance effects in dynamic echo-planar and spiral functional imaging. Magn Reson Med. Feb 2002;47(2): Liu TT, Wong EC. A signal processing model for arterial spin labeling functional MRI. Neuroimage. Jan ;24(1): Ye FQ, Frank JA, Weinberger DR, McLaughlin AC. Noise reduction in 3D perfusion imaging by attenuating the static signal in arterial spin tagging (ASSIST). Magn Reson Med. Jul 2000;44(1): Miranda MJ, Olofsson K, Sidaros K. Noninvasive measurements of regional cerebral perfusion in preterm and term neonates by magnetic resonance arterial spin labeling. Pediatr Res. Sep 2006;60(3): Deibler AR, Pollock JM, Kraft RA, Tan H, Burdette JH, Maldjian JA. Arterial Spin-Labeling in Routine Clinical Practice, Part 1: Technique and Artifacts. AJNR Am J Neuroradiol. Mar Deibler AR, Pollock JM, Kraft RA, Tan H, Burdette JH, Maldjian JA. Arterial Spin-Labeling in Routine Clinical Practice, Part 2: Hypoperfusion Patterns. AJNR Am J Neuroradiol. Mar Deibler AR, Pollock JM, Kraft RA, Tan H, Burdette JH, Maldjian JA. Arterial Spin-Labeling in Routine Clinical Practice, Part 3: Hyperperfusion Patterns. AJNR Am J Neuroradiol. Mar

56 11. Maldjian JA, Laurienti PJ, Burdette JH, Kraft RA. Clinical implementation of spin-tag perfusion magnetic resonance imaging. J Comput Assist Tomogr. May- Jun 2008;32(3): Wong EC, Buxton RB, Frank LR. Quantitative imaging of perfusion using a single subtraction (QUIPSS and QUIPSS II). Magn Reson Med. May 1998;39(5): Luh WM, Wong EC, Bandettini PA, Hyde JS. QUIPSS II with thin-slice TI1 periodic saturation: a method for improving accuracy of quantitative perfusion imaging using pulsed arterial spin labeling. Magn Reson Med. Jun 1999;41(6): Kim SG. Quantification of relative cerebral blood flow change by flow-sensitive alternating inversion recovery (FAIR) technique: application to functional mapping. Magn Reson Med. Sep 1995;34(3): Buxton RB, Frank LR, Wong EC, Siewert B, Warach S, Edelman RR. A general kinetic model for quantitative perfusion imaging with arterial spin labeling. Magn Reson Med. Sep 1998;40(3): Stanisz GJ, Odrobina EE, Pun J, et al. T1, T2 relaxation and magnetization transfer in tissue at 3T. Magn Reson Med. Sep 2005;54(3): Liu TT, Brown GG. Measurement of cerebral perfusion with arterial spin labeling: Part 1. Methods. J Int Neuropsychol Soc. May 2007;13(3): McConnell MV, Khasgiwala VC, Savord BJ, et al. Comparison of respiratory suppression methods and navigator locations for MR coronary angiography. AJR Am J Roentgenol. May 1997;168(5):

57 CHAPTER III 3D GRASE PROPELLER: IMPROVED IMAGE ACQUISITION TECHNIQUE FOR ARTERIAL SPIN LABELING PERFUSION IMAGING Huan Tan, W. Scott Hoge, Craig A. Hamilton, Matthias Günther, Robert A. Kraft The following manuscript was in press in Magnetic Resonance in Medicine. Stylistic variations are due to the requirement of the publisher. H. Tan developed the reconstruction code, performed data analysis, and prepared the manuscript. Dr. R.A. Kraft developed the pulse sequence program and conducted experiments with H. Tan. Dr. W.S. Hoge and Dr. C.A. Hamilton assisted with image reconstruction, and Dr. M. Günther assisted with pulse sequence development. Dr. R.A. Kraft acted in an advisory and editorial capacity during manuscript preparation. 43

58 ABSTRACT Arterial spin labeling (ASL) is a non-invasive technique that can quantitatively measure cerebral blood flow (CBF). While traditionally ASL employs 2D EPI or spiral acquisition trajectories, single-shot 3D GRASE is gaining popularity in ASL due to inherent SNR advantage and spatial coverage. However, a major limitation of 3D GRASE is throughplane blurring caused by T2 decay. A novel technique combining 3D GRASE and a PROPELLER trajectory (3DGP) is presented to minimize through-plane blurring without sacrificing perfusion sensitivity or increasing total scan time. Full brain perfusion images were acquired at a 3x3x5mm3 nominal voxel size with Q2TIPS-FAIR as the ASL preparation sequence. Data from 5 healthy subjects was acquired on a GE 1.5T scanner in less than 4 minutes per subject. While showing good agreement in CBF quantification with 3D GRASE, 3DGP demonstrated reduced through-plane blurring, improved anatomical details, high repeatability and robustness against motion, making it suitable for routine clinical use. Key Words: Arterial Spin Labeling, PROPELLER, 3D GRASE 44

59 INTRODUCTION Arterial spin labeling (ASL) is an MRI technique that uses water molecules in the blood as an endogenous tracer to measure cerebral blood flow (CBF). It is highly effective in detecting brain lesions, cerebrovascular diseases, and brain tumors 1. Compared to traditional contrast-bolus based methods, ASL is completely non-invasive and repeatable, and offers direct measurement of quantitative perfusion values. The noninvasive nature of ASL is especially important for patients with conditions such as renal failure, or in pediatric patients where the use of external tracers may be restricted 2. ASL has an inherently low perfusion signal-to-noise ratio (SNR) and signal averaging is necessary to obtain perfusion weighted images with sufficient SNR. As fast imaging techniques, 2D Echo Planar Imaging (EPI) and spiral imaging are commonly used for ASL acquisition to increase slice coverage. However, multi-slice 2D acquisitions result in multiple inflow times that may underestimate perfusion and cause quantification errors in certain regions. Also, both EPI and spiral imaging are highly sensitive to magnetic field inhomogeneity and susceptibility, leading to signal loss and geometric distortions. To minimize those artifacts, ASL images acquired with EPI and spiral are typically acquired with low spatial resolution. Nevertheless, higher spatial resolutions are often desirable to reduce partial volume effects and improve localization of abnormalities. Alternatively, 3D acquisition techniques have been developed for ASL imaging. Techniques such as 3D EPI 3, 4 and 3D GRASE (Gradient Echo And Spin Echo) 5 have been shown to have higher perfusion sensitivity and better spatial coverage of the brain than 2D EPI. As a gradient echo method, 3D EPI generally has better slice coverage than 3D GRASE. However, 3D GRASE has lower sensitivity to field inhomogeneity and 45

60 magnetic susceptibility as a spin echo method. The robustness of 3D GRASE to magnetic field inhomogeneity is especially important for reducing image artifacts with non- Cartesian trajectories, such as PROPELLER 6. These qualities make 3D GRASE an attractive choice as an ASL imaging acquisition sequence. One drawback has limited the widespread use of 3D GRASE. As a single-shot 3D technique, raw image data is collected after a single excitation with a long acquisition window, which results in severe through-plane blurring due to T 2 decay. One apparent solution to reduce the blurring is to shorten the acquisition window by shortening the echo train length (ETL). Two common methods for shortening the ETL are parallel imaging and multi-shot acquisitions. Parallel imaging can reduce the ETL but at the expense of SNR loss. Although the loss of SNR is partially compensated by the shorter echo time (TE) achievable with parallel imaging, at higher acceleration rates (> 2x), the SNR loss is significant 7. Multi-shot methods are another approach for reducing the ETL by splitting the full data acquisition among multiple excitations. The disadvantage of the multi-shot method is the increase in scan time and susceptibility to image artifacts induced by motion. With a standard Cartesian trajectory that is commonly used in multi-shot methods, it is difficult to compensate for motion that occurs between shots. One multi-shot technique that can reduce ETL and correct for motion error is PROPELLER (Periodically Rotated Overlapping ParallEL Lines with Enhanced Reconstruction) 6, 8. Unlike the Cartesian trajectory, data acquisition in PROPELLER consists of a series of rectangular trajectories (known as blades) rotated about the center of k-space. Due to this unique rotational trajectory, one advantage of PROPELLER is the 46

61 ability to perform self-referenced motion correction 6. Comparable motion correction is difficult in multi-shot methods with Cartesian trajectory without acquiring additional navigator echoes which further reduces acquisition efficiency. For long axis PROPELLER where the blade width (set by the ETL) is small compared to the blade length (the number of frequency encodes), the resolution of the final reconstructed image is determined solely by the blade length and the field of view (FOV). Hence, the original spatial resolution is fully preserved regardless of the width of the individual blade. A disadvantage of PROPELLER is an increase in scan time to fully sample k-space as a multi-shot technique. However, this disadvantage is inconsequential for ASL imaging where signal averaging is necessary to achieve adequate perfusion SNR. We hypothesize that it is more advantageous to sample k-space in a non-cartesian, multi-shot fashion with PROPELLER than simply signal averaging identical acquisitions repeatedly. In this paper, we incorporate the 3D GRASE readout sequence with PROPELLER (3DGP) to improve the spatial resolution of ASL imaging while minimizing T 2 blurring and off-resonance effects. We compare the performances of 3D GRASE and 3DGP for measuring gray matter perfusion. The repeatability, CBF quantification and the perfusion SNR of 3DGP are examined in detail. MATERIALS AND METHODS Pulsed ASL (PASL) Implementation ASL labeling was achieved using Q2TIPS-FAIR 9, 10, a pulsed ASL implementation. Saturation bands were applied along both sides of the imaging slab to minimize aliasing artifacts in the slice encoding direction, as well as to suppress the 47

62 intravascular signals carried by both arterial and venous inflows. An inversion time of 1500ms was chosen to allow blood to fully exchange with the tissue. Additional background suppression pulses were applied at 539 ms and 1345ms to null the stationary tissue signal to further improve perfusion SNR 11. PROPELLER Acquisition For 3DGP, a number of rectangular volumes (known as bricks 12 ) are acquired at different rotation angles relative to the central k z -axis (Figure 9a). In our study, each brick was sampled with a single-shot 3D GRASE readout module (Figure 9b). 16 bricks evenly distributed from 0 to 360 were acquired per PROPELLER image by an incremental angle of The number of bricks acquired for each PROPELLER image satisfied both the Nyquist sampling criterion previously reported 6 and ensured adequate perfusion SNR. A total of four PROPELLER images were reconstructed: a control image and a label image, each acquired with a positive and a negative frequency encoding gradient. The toggling of the frequency encoding gradient was necessary to remove Nyquist ghosts using the self-referenced correction method GESTE 13. The total scan time was T scan time = TR N Blade N ASL N GESTE where TR was the repetition time (3500ms), N Blade = 16 was the number of bricks per image, N ASL = 2 was the number of perfusion states, and N GESTE = 2 was the number of encodes required for GESTE. The total scan time was 3 minutes 44 seconds. The acquisition pattern for PROPELLER bricks is in the following order: brick (control, GESTE+) at 0º, brick (label, GESTE+) at 0º, brick (control, GESTE ) at 0º, brick (label, GESTE ) at 0º. The order of acquisition then repeats for another 15 PROPELLER bricks evenly spaced from 22.5º to 360º. 48

63 Figure 9. (a) The schematic illustration of k-space trajectory for 2D PROPELLER (left) and 3D GRASE PROPELLER (right). Rather than a series of rotating blades in the 2D PROPELLER trajectory, 3D GRASE PROPELLER consists of a series of rotating bricks about the central z- axis. Note the k-space is under-sampled for illustration purpose. Only 4 of 16 blades are shown. (b) 3D GRASE pulse diagram for acquiring each brick. Image Post-Processing A 1D Fourier transform was first applied along k z for each raw brick, yielding the equivalent of 2D PROPELLER blades from multiple slices. GESTE 13 was applied to remove Nyquist ghosts in each blade, followed by the standard PROPELLER reconstruction 6. In the current work, motion correction (rotational and translational) was performed only along the in-plane direction. Through-plane motion was dealt with indirectly with correlation weighting 6. Note that the same motion correction reference image was used for both control and label blades during the in-plane motion correction. This prevented misalignments between the final control and label images to avoid potential subtraction errors. The corrected blades were gridded onto a Cartesian coordinate space with uniform density compensation 14. The final image for each slice was obtained by performing a 2D Fourier transform on the gridded and combined k-space data. 49

64 Figure 10. Diagram of 3DGP reconstruction process. Each of the four PROPELLER images was reconstructed separately. The magnitude images corresponding to each GESTE encode were signal averaged to generate a single pair of control and label images. Perfusion weighted images (PWI) were then calculated by subtracting the label image from the control image. CBF quantification was done using the General Kinetic Model 10, 15. The entire reconstruction process is illustrated in Figure 10. Experiments Perfusion images were acquired using both 3DGP and 3D GRASE in the axial plane. In both techniques, the prescribed imaging slab was 90 mm thick consisting of 18 50

65 slices with 5 millimeter thick slices. Excitation was achieved with a 90 degree 7 lobe RF spatial spectral pulse. To further reduce blurring, partial Fourier encoding with 72% coverage was used to reduce the number of slice encodings to 13. GESTE encoding was used in both techniques for Nyquist ghost removal. The TE was minimized to optimize perfusion sensitivity by employing a center out trajectory along the slice encoding direction. The total imaging time for 3DGP and 3D GRASE was identical (3 minutes 44 seconds). For 3DGP, 20 phase encodes and 96 frequency encodes were prescribed per brick. The in-plane FOV was 288 x 288 mm, thus achieving a voxel size of 3 x 3 x 5 mm 3. A smaller FOV may be prescribed to achieve the same in-plane resolution; however, this has no effect on TE and echo spacing because a larger gradient amplitude with longer ramp times is required for the smaller FOV. A total of 16 bricks were acquired for each control/label image. Other imaging parameters included TE / TR = 17.5 / 3500 ms, total acquisition window duration = 131ms. In a separate scan, a 3DGP M 0 weighted image was acquired without the ASL preparation for CBF quantification with 3 bricks and TR = 10s. For 3D GRASE, the image matrix size was 96 x 64 with a field of view of 288 x 192mm, matching the voxel size of the 3DGP acquisition. Other imaging parameters included TE / TR = 41.2 / 3500 ms, total acquisition window duration = 379 ms. The number of control/label pairs for signal averaging was prescribed to match the 3DGP prescription exactly. The difference in the total acquisition window duration between 3D GRASE and 3DGP is solely due to the number of phase encoding lines. Reconstructed 3D GRASE images were realigned using SPM2 (Wellcome Trust Center for Neuroimage, 51

66 London, UK) to remove any motion errors. A 3D GRASE M 0 weighted image with TR = 10s was acquired for CBF quantification in a separate scan. In addition to perfusion sequences, high resolution anatomical images were acquired using 3D SPGR for tissue segmentation. T1 maps were acquired using the DESPOT1 16 sequence for CBF quantification without additional B 1 correction. A total of 5 healthy volunteers of age from 24 to 30 were recruited under the supervision of the Institutional Review Board, and informed consent was obtained from each subject. All invivo experiments were carried out on a GE 1.5T TwinSpeed scanner (GE Healthcare, Milwaukee, WI) with an 8 channel phased array receive-only head coil (Invivo Devices, Gainesville, FL) for data collection. The pulse sequence and image reconstruction software were developed internally. Imaging reconstruction software was written in Matlab (Mathworks, Natick, MA) utilizing functions from the National Center for Image Guided Therapy (NCIGT) Fast Imaging Library 17. Repeatability, SNR and SNR Per Pixel Calculation To test repeatability and calculate SNR, both 3DGP and 3D GRASE acquisitions were repeated without repositioning the subject. The repeatability was tested by calculating the coefficient of repeatability, following the methods of Bland and Altman 18. SNR was calculated from the repeated acquisition 19. The noise for the SNR calculation was estimated from the difference image of the two repeated PWIs. The mean signal and the standard deviation of the noise were obtained from the gray matter region of the PWIs and the difference image, respectively. Both gray matter and white matter 52

67 were identified from tissue segmentation maps, obtained from the high resolution anatomical images processed by SPM2. Image reconstruction algorithms can affect the final SNR of the reconstructed image. Since 3D GRASE and 3DGP use very different image reconstruction algorithms, the average raw SNR per pixel was calculated for both methods by integrating the raw MRI signal as described by Liang and Lauterbaur (Equation ). The standard deviation of the noise for this calculation was measured from a single acquisition with all of the RF pulses turned off. RESULTS Figure 11 shows the absolute CBF maps of slices 2 through 16 obtained in one of the five subjects from 3DGP and 3D GRASE acquisitions. Both sets of images showed full brain coverage and SNR of good diagnostic quality. Through-plane blurring artifacts were reduced in 3DGP comparing to 3D GRASE, most apparent in the superior and inferior slices. The improvements on blurring can be better seen in the reformatted coronal and sagittal views, shown in Figure 12. Anatomical details such as the thalamus and the caudate-putamen were better revealed in 3DGP (indicated by the arrows in row one of Figure 12) while the same structures were hard to identify in corresponding 3D GRASE slices. 53

68 Figure 11. A comparison of CBF maps acquired with 3DGP and 3D GRASE from one of the five subjects. Fifteen slices of a total of 18 are shown at inflow times TI = 1500ms. Severe blurring can be observed in the most superior and inferior slices of 3D GRASE, indicated by the arrows. 54

69 Figure 12. Reformatted views of ASL images acquired in the axial plane. Severe blurring can be seen in the 3D GRASE images along the slice encoding direction. The blurring effect was significantly reduced in the 3DGP images. In-plane image details, such as thalamus and caudate-putamen indicated by the arrows, can be easily identified in the 3DGP images whereas the same region was less recognizable in the 3D GRASE images. The mean measured CBF values for 3D GRASE were 50 ± 14 ml/100g/min for gray matter and 23 ± 7 ml/100g/min for white matter. The average calculated CBF values for 3DGP were 52 ± 16 ml/100g/min for gray matter and 19 ± 7 ml/100g/min for white matter. The difference in white matter perfusion between 3DGP and 3D GRASE was significant according to a paired t-test (p < ) while it was not for gray matter 55

70 (p > 0.28). The increase in white matter perfusion in 3D GRASE might be the result of severe through-plane blurring where the perfusion values were falsely increased by the contributions from other slices. This was also reported in the mean gray matter and white matter perfusion ratio: 2.7 for 3DGP and 2 for 3D GRASE. The individual CBF values for all subjects are shown in Table 4. Table 4. Mean CBF measurements for all subjects. Subject Type Gray Matter CBF White Matter CBF (ml/100g/min) (ml/100g/min) Ratio 1 3DGP ± ± GRASE ± ± DGP ± ± GRASE ± ± DGP ± ± GRASE ± ± DGP ± ± GRASE ± ± DGP ± ± GRASE ± ± Both techniques showed good repeatability in perfusion measurements. The maximum difference in CBF between repeated scans was less than 5 ml/100g/min for whole gray matter and white matter measurements. The coefficients of repeatability of 3D GRASE and 3DGP were 5.6 and 6.2 ml/100g/min for gray matter, and 3.6 and 3.3 ml/100g/min for white matter, respectively. This means the difference between two perfusion measurements for the same subject is expected to be less than the calculated coefficients of repeatability in 95% of pairs of observations. The SNR results across all 56

71 subjects were 7.2 ± 1.82 for 3D GRASE and 4.9 ± 1.37 for 3DGP. The raw SNR per pixel values for all subjects were 9.4 ± 0.63 for 3D GRASE and 28.8 ± 1.96 for 3DGP. A detailed SNR analysis is included in the discussion section. DISCUSSION This paper described the use of a non-cartesian trajectory PROPELLER to minimize through-plane blurring in a 3D GRASE readout sequence for ASL imaging. PROPELLER was chosen due to a shorter ETL and its robustness against motion. This new technique, 3DGP, was demonstrated to show improved overall image quality in terms of reduced through-plane blurring, fewer off-resonance artifacts and improved image resolution. Compared to a single-shot 3D GRASE readout, each 3DGP brick is less susceptible to off-resonance effects as a result of the reduced ETL. PROPELLER has demonstrated strong insensitivity to local off-resonance related distortion 21, and the final reconstructed 3DGP image exhibited few image distortions. In comparison, off-resonance induced distortion in 3D GRASE was more apparent and remained consistent across the time series. As a result of averaging, the final perfusion image was presented with the same distortion; whereas such artifact was absent in the 3DGP images. Since the in-plane resolution of 3DGP depends primarily on the blade length, the image resolution can be improved by increasing the resolution along the frequency encode direction. This imposes only a minor increase in TE if the Nyquist sampling criterion is satisfied. However, if the Nyquist sampling criterion is not satisfied, the total number of PROPELLER bricks and/or the ETL need to be increased accordingly. 57

72 Increasing the number of PROPELLER bricks will increase scan time, while increasing the ETL will increase TE and geometrical distortion. Therefore, it is desirable to acquire 3DGP images with the minimum ETL possible while maximizing the number of the bricks acquired for a given scan time. Given that a short ETL is desired, a minimum ETL of 16 is needed for adequate motion correction 22. However, thicker blades ensure the full capture of the k-space center to prevent signal loss during rotation and provide better motion correction 8. In our experience, an ETL of 20 is a good compromise between signal capture and motion correction. In perfusion imaging, signal averaging is necessary to achieve perfusion images with adequate SNR. One often needs to acquire more bricks than the Nyquist sampling criterion to achieve adequate SNR, thus providing the opportunity to adjust the ETL without increasing scan time. Nonetheless, this makes 3DGP a good candidate for high resolution ASL imaging. GESTE is chosen for Nyquist ghost correction due to its robustness against system instability and higher efficiency than reference scan based techniques 13. In the presence of background suppression where tissue signal was reduced by 90%, the magnitude image is still dominated by residual tissue signal, and the performance of GESTE is not affected. GESTE encoding is executed separately for the control and label image in the current implementation. However, it may be possible to merge GESTE and ASL encoding by coupling the readout gradient polarity change with the blood magnetization labeling state, which would eliminate the need for acquiring two GESTE images for each control/label image. This is currently under investigation. The SNR measurements of 3D GRASE for the perfusion weighted images were consistently higher than 3DGP according to in-vivo experiments. This was unexpected 58

73 for three reasons. First, 3DGP has a shorter TE than 3D GRASE, which minimizes signal loss due to T 2 relaxation. Second, the time between refocusing RF pulses is shorter for 3DGP (about 1/3 of the time of 3D GRASE), which minimizes signal loss due to magnetic field inhomogeneities and reduces T 2 blurring. Third, the long acquisition duration required by single-shot 3D GRASE allows more noise to be introduced into the final image than signal due to T 2 decay 23. While longer acquisition duration of 3D GRASE does result in more through-plane blurring, the SNR advantage of 3D GRASE cannot be attributed to this. For a concentric-out trajectory along k z, the T 2 relaxation attenuates signal acquired at higher spatial frequencies but does not affect the noise. This is different from applying an exponential filter to the k-space data during post-processing, where both high frequency data and noise in k-space are attenuated leading to an overall increase in SNR. We suspected that the SNR loss of 3DGP compared to 3D GRASE was occurring during image reconstruction. To further investigate this hypothesis, the average image SNR per pixel was calculated directly from the raw k-space data before image reconstruction 20. This revealed that the SNR of 3DGP images was 3 times greater than 3D GRASE (average SNR per pixel for 3DGP: 28.8, 3D GRASE: 9.4) despite the fact that 3D GRASE acquires 3 times as much data. The two SNR measurements (before and after reconstruction) revealed the signal loss of 3DGP was occurring during image reconstruction. We attributed this loss of SNR to geometrical and intensity distortions from magnetic field inhomogeneities during PROPELLER acquisition. As the 3DGP brick rotates, the geometric distortion changes direction for each brick. Combining these individual bricks without correcting for the distortion would result in incoherent signal 59

74 averaging leading to signal loss. Although the geometrical distortion in 3D GRASE is larger than 3DGP due to the longer ETL, it has minimal effect on SNR since the distortion pattern is consistent throughout the entire experiment and is averaged coherently. We believe the SNR of 3DGP can be recovered by correcting for those residual off-resonance errors as demonstrated by Skare and Andersson 24, which we are currently investigating. CONCLUSIONS In this paper, we have shown that through-plane blurring present in 3D GRASE can be significantly reduced by incorporating a PROPELLER trajectory for ASL imaging. This technique, 3D GRASE PROPELLER or 3DGP, has demonstrated robustness against motion, off-resonance effects, and improved in-plane and through-plane image resolution, while inheriting the usual merits of 3D GRASE. By incorporating additional offresonance correction, 3DGP has a great potential for higher resolution ASL imaging at higher field strength. ACKNOWLEDGMENT The authors thank the Center of Biomolecular Imaging (CBI) at Wake Forest University School of Medicine (Winston-Salem, NC) for unfettered access to the MRI scanner and support from the Functional Neuroimaging Lab at Brigham and Women's Hospital (Boston, MA). The authors also thank Dr. Brian Hargreaves for publicly posting his gridding code and Dr. Jing Yuan at the BWH/National Center for Image Guided Therapy for providing the code for the spatial spectral pulses. This work was supported in 60

75 part by NIH U41 RR and 5R01AA This work was also in part funded by the German Ministry of Research and Education (BMBF) by grant 01EV

76 REFERENCES 1. Calamante F, Thomas DL, Pell GS, Wiersma J, Turner R. Measuring cerebral blood flow using magnetic resonance imaging techniques. J Cereb Blood Flow Metab. Jul 1999;19(7): Pollock JM, Tan H, Kraft RA, Whitlow CT, Burdette JH, Maldjian JA. Arterial spin-labeled MR perfusion imaging: clinical applications. Magn Reson Imaging Clin N Am. May 2009;17(2): Gai ND, Talagala SL, Golay X, Hoogenraad FG, Butman JA. Evaluation of 3D- EPI PULSAR with and without Background Suppression Inversion Recovery Pulse. Paper presented at: Proc. Intl. Soc. Mag. Reson. Med., 2007; Berlin, Germany. 4. Talagala SL, Ye FQ, Ledden PJ, Chesnick S. Whole-brain 3D perfusion MRI at 3.0 T using CASL with a separate labeling coil. Magn Reson Med. Jul 2004;52(1): Gunther M, Oshio K, Feinberg DA. Single-shot 3D imaging techniques improve arterial spin labeling perfusion measurements. Magn Reson Med. Aug 2005;54(2): Pipe JG. Motion correction with PROPELLER MRI: application to head motion and free-breathing cardiac imaging. Magn Reson Med. Nov 1999;42(5): Wang Z, Wang J, Connick TJ, Wetmore GS, Detre JA. Continuous ASL (CASL) perfusion MRI with an array coil and parallel imaging at 3T. Magn Reson Med. Sep 2005;54(3): Pipe JG, Zwart N. Turboprop: improved PROPELLER imaging. Magn Reson Med. Feb 2006;55(2): Kim SG. Quantification of relative cerebral blood flow change by flow-sensitive alternating inversion recovery (FAIR) technique: application to functional mapping. Magn Reson Med. Sep 1995;34(3):

77 10. Luh WM, Wong EC, Bandettini PA, Hyde JS. QUIPSS II with thin-slice TI1 periodic saturation: a method for improving accuracy of quantitative perfusion imaging using pulsed arterial spin labeling. Magn Reson Med. Jun 1999;41(6): Ye FQ, Frank JA, Weinberger DR, McLaughlin AC. Noise reduction in 3D perfusion imaging by attenuating the static signal in arterial spin tagging (ASSIST). Magn Reson Med. Jul 2000;44(1): Holdsworth SJ, Skare S, Nordell A, Newbould RD, Bammer R. 3D SAP-EPI for Self-Navigated T1w Spoiled Gradient Echo Imaging. Paper presented at: ISMRM, 2008; Toronto, Canada. 13. Hoge WS, Tan H, Kraft RA. Robust EPI Nyquist ghost elimination via spatial and temporal encoding. Magn Reson Med. Dec 2010;64(6): Jackson J, Meyer, CH, Nishimura, DG, Macovski, A. Selection fo a convolution function for Fourier inversion using gridding. IEEE Transactions on Medical Imaging. 1991;10(3): Buxton RB, Frank LR, Wong EC, Siewert B, Warach S, Edelman RR. A general kinetic model for quantitative perfusion imaging with arterial spin labeling. Magn Reson Med. Sep 1998;40(3): Deoni SC, Rutt BK, Peters TM. Rapid combined T1 and T2 mapping using gradient recalled acquisition in the steady state. Magn Reson Med. Mar 2003;49(3): NCIG Fast Imaging Library [computer program]. Version. Boston, MA; Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet. Feb ;1(8476): Sijbers J, den Dekker AJ, Van Audekerke J, Verhoye M, Van Dyck D. Estimation of the noise in magnitude MR images. Magn Reson Imaging. 1998;16(1): Liang Z-P, Lauterbur PC. Principles of magnetic resonance imaging: a signal processing perspective. New York: Wiley-IEEE Press;

78 21. Wang FN, Huang TY, Lin FH, et al. PROPELLER EPI: an MRI technique suitable for diffusion tensor imaging at high field strength with reduced geometric distortions. Magn Reson Med. Nov 2005;54(5): Tamhane AA, Arfanakis K. Motion correction in periodically-rotated overlapping parallel lines with enhanced reconstruction (PROPELLER) and turboprop MRI. Magn Reson Med. Jul 2009;62(1): Mugler J, Brookeman J. The optimum data sampling period for maximum signalto-noise ratio in MR imaging. Rev Magn Reson Med. 1988;3(1): Skare S, Andersson J. Calibration free distortion correction for PROPELLER EPI. Proceedings 16th ISMRM Scientific Meeting. 2008:

79 CHAPTER IV ROBUST EPI NYQUIST GHOST ELIMINATION VIA SPATIAL AND TEMPORAL ENCODING (EPI-GESTE) W. Scott Hoge, Huan Tan, Robert A. Kraft The following manuscript was published in Magnetic Resonance in Medicine, volume 64, pages Stylistic variations are due to the requirement of the publisher. Dr. W.S. Hoge and H. Tan developed the reconstruction technique and Dr. R.A. Kraft developed the pulse sequence. Dr. W.S. Hoge, H. Tan and Dr. Kraft performed experiments and analysis on phantom and in-vivo data. The manuscript was prepared by Dr. W.S. Hoge. Editing was performed by H. Tan and Dr. R.A. Kraft. 65

80 ABSTRACT Nyquist ghosts are an inherent artifact in EPI acquisitions. An approach to robustly eliminate Nyquist ghosts is presented that integrates two previous Nyquist ghost correction techniques: temporal domain encoding (Phase Labeling for Additional Coordinate Encoding: PLACE) and spatial domain encoding (Phased Array Ghost Elimination: PAGE). Temporal encoding modulates the EPI acquisition trajectory from frame to frame, enabling one to interleave data to remove inconsistencies that occur between sampling on positive and negative gradient readouts. With PLACE, one can coherently combine the interleaved data to cancel residual Nyquist ghosts. If the level of ghosting varies significantly from image to image, however, the signal cancellation that occurs with PLACE can adversely affect SNR-sensitive applications such as perfusion imaging with Arterial Spin Labeling (ASL). This work proposes integrating PLACE into a PAGE-based reconstruction process to yield significantly better Nyquist ghost correction that is more robust than PLACE or PAGE alone. The robustness of this method is demonstrated in the presence of magnetic field drift with an in-vivo ASL perfusion experiment. Key words: EPI artifacts, Nyquist ghost correction, GRAPPA, ASL Perfusion Imaging 66

81 INTRODUCTION Modern neuroimaging methods that require fast data acquisition often rely on echo planar imaging (EPI). EPI acquisitions achieve high temporal resolution by sampling multiple lines of k-space data after each RF excitation pulse. Sampling on a Cartesian grid requires multiple readout gradient pulses of alternating polarity combined with small phase-encode blips 1. When data acquired during positive readout gradients, +G x, is not properly aligned with data sampled during negative readout gradients, G x, Nyquist ghosts can occur 2. These ghosts are characterized as faint copies of the imaged material, shifted from the true image by one-half the field-of-view (FOV). There have been many methods to correct for Nyquist ghosts. Early on, Ahn and Cho 3, proposed a 1D phase correction method to align the +G x and G x echoes. This method employs a reference scan with the EPI blip amplitudes set to zero. The reference data is then used to estimate the constant and linear phase errors between data sampled on +G x versus data sampled using G x. However, the dependence on reference data acquired before image acquisition leaves the image reconstruction susceptible to variations that may occur during the experiment. In our experience, when reference data is used to correct Nyquist ghosts, even small phase variations can leave significant residual Nyquist ghosts. Dynamic correction methods, such as proposed by Jesmanowicz et. al. 4, can partially compensate for temporal variations in a series acquisition. These methods typically acquire additional k-space lines at each temporal image point, to determine the constant and linear phase terms for each image. This can remove the dependence of the reference scan, but at the cost of a slightly longer ETL. 67

82 It was later noted that double-sampled EPI 5, which measures each k-space line twice in quick succession once each for +G x and G x completely eliminates Nyquist ghosts. Each positive/negative readout data set satisfies the Nyquist criterion, and separating the sets removes the effects from data incoherence between them. This approach has the unfortunate effect of doubling the echo train length (ETL), however, which exaggerates EPI geometric distortion artifacts caused by magnetic field inhomogeneity. Other methods have been proposed that do not increase the ETL. For example, Buonocore and Zhu 6 demonstrated that phase error correction maps could be generated directly from image domain analysis of the acquired data. Chen and Wyrwicz 7 expanded this approach using fully-encoded reference scans. However, these methods have been shown to be less effective in oblique scan planes 8. A dependence on reference data, and complications from phase wrapping effects and in estimating the map in regions of low signal, also limit its applicability 8. More recent methods employ spatial and/or temporal domain encoding to correct for Nyquist ghosts. The phase labeling for additional coordinate encoding (PLACE) method 9 provides exceptional ghost suppression. In PLACE, the temporal encoding is performed by shifting the acquired k-space grid on alternating acquisitions. For ghost elimination, a shift of 1Δ is performed, and data from +G x and G x readout lines is k y interleaved to form two images, each associated with one readout gradient polarity. Any phase difference between these images, which arises due to a shift in the underlying k- space sampling grid, is then corrected before the images are added coherently. Residual ghosts in each image should have opposite phase. Thus, a coherent combination 68

83 effectively cancels any ghosts that may appear in the interleaved data. As with all interleaved data methods, however, the cost is lower temporal resolution in the image series. Methods that employ spatial domain encoding for phased array ghost elimination, e.g. (PAGE) 10, leverage parallel imaging methods such as SENSE 11 to correct Nyquist ghosting. The authors of PAGE recognized that when data acquired after a single excitation are separated into separate +G x and G x sets, sampling incoherence is removed and a 2x acceleration is introduced in each data set. These sets can each be reconstructed using SENSE, via coil sensitivity estimates derived from low temporal resolution data (as in TSENSE 12 ). In PAGE, images so formed are combined in a coherent fashion using a weighted summation derived from the point spread function (PSF). In cases where the PSF needs to be estimated, the weighting coefficients can be derived from a timeaveraged covariance matrix for each pixel. The latency in this method is significant, however, as both the coil sensitivity and PSF estimation requires data from multiple imaging time points. This method was extended to a low-latency real-time PAGE scenario by Kim, et. al., in 8. Here, the authors alter the acquisition to alternate the readout gradient polarity order on each frame. That is, data for odd-numbered time points would be acquired with the readout polarity ordered {+,, +,, } while even-numbered time points would be acquired with {+,, +,, } readout gradient ordering. In the real-time PAGE method, data from two adjacent frames are interleaved to estimate coil-sensitivities one each for +G x and G x data, significantly reducing the PAGE latency costs. These coil sensitivities are then used with SENSE to reconstruct images from the separated +G x and G x data for 69

84 each time point, with the final image formed using a non-coherent combination (rootsum-of-squares) of the two images. While PAGE and PLACE are better at suppressing Nyquist ghosts than more traditional methods, unresolved issues remain in both methods. PAGE suffers from long latency, from both the estimation of the coil sensitivity maps and from the solution of an inverse problem to identify the coherent image summation weights. Real-time PAGE corrects the latency issue, but rejects the coherent summation approach in place of the faster root-sum-of-squares operation. While this method is simple and effective, it is unsuitable in applications where phase information is needed. Finally, in PLACE, the effect of ghost cancellation can adversely affect SNR sensitive applications such as Arterial Spin Labeling (ASL) perfusion imaging. Nyquist ghosts arise from displaced signal information. Rather than correcting the location of this shifted information, PLACE effectively cancels the displaced signal, including areas where ghosts overlap tissue of interest. Thus, temporal neighbors in an image series can exhibit significant intensity variations if the ghosting level varies from image to image before the PLACE correction. For ASL imaging, where the perfusion signal is found by calculating image differences, PLACE can introduce subtraction errors. Our goal was to develop a self-referenced Nyquist ghost elimination method that addresses each of these problems. A self-referenced method was sought to remove the dependence on reference data in either a prescan step or a longer EPI echo train. A low latency method was sought to be insensitive to variations from both unintentional changes such as those caused by gradient coil heating or patient breathing and intentional changes in the acquisition, including real time FOV modifications and/or 70

85 PROPELLER acquisitions 13. The method also needs to be robust and repeatable to prevent visible errors in the temporal difference domain. We have found this to be extremely important in low SNR applications such as ASL, where ghost correction errors can obscure the information of interest. The method we present here achieves these goals by integrating the spatial (realtime PAGE) and temporal (PLACE) encoding methods presented previously. While the PAGE method includes the option of coherent phase image combinations, the stated benefit of such a method is to maintain the signal phase the goal of residual ghost suppression was not considered. In contrast, we specifically employ PLACE to eliminate ghosts in the data used for self-referenced parallel imaging coefficient calibration. As we show in the results, this enables our EPI Ghost Elimination via Spatial and Temporal Encoding (EPI-GESTE) method to be more robust to acquisition variability with significantly improved ghost suppression. METHODS EPI Acquisition Design The foundation of our method is to employ both spatial and temporal domain encoding in EPI acquisitions. Spatial encoding is achieved using a multiple channel receiver coil array. Temporal encoding can be achieved by either shifting the EPI acquisition in k-space by 1Δ along the phase encoding dimension 9, or by alternating k y the readout gradient polarity on successive frames 8. The goal with either method is to ensure that data at a particular phase encode line that is sampled in one frame with a positive readout gradient, +G x, is sampled by a negative readout gradient, G x, in 71

86 neighboring frames. This enables one to interleave data from temporal neighbors, as shown in the alternating readout polarity method illustrated in the far-left column of Fig. 13. Figure 13. A diagram showing our proposed data acquisition and image reconstruction steps for unaccelerated multicoil EPI acquisitions. The method is similar to the method by Kim et al. 8, with the exception of PLACE employed in the regions shown by the gray boxes. Each dashed box represents data from one coil of a multi-coil array. Reconstruction Methods EPI-GESTE: Our strategy to reconstruct multi-coil EPI acquisitions is shown in the remainder of Fig. 13. Starting with k-space data from two temporally encoded image frames, kt ( 1) and kt ( ), we interleave the positive readout data to form one set of k- space data, k ' p, and then interleave the negative readout data to form a second set of k- space data, k ' n. Notably, the images associated with each of these data sets, I ' p and ' I n, 72

87 should have minimal Nyquist ghosts, as most sampling inconsistencies between +G x and G x should be removed by interleaving. Variations may in fact occur between the sampling of kt ( 1) and kt ( ), which lead to residual Nyquist ghosts in images from interleaved data. An example is shown in Fig. 14. As noted by Xiang and Ye 9, however, ghost artifacts that appear in images from interleaved temporally encoded EPI data, e.g. I ' p and ' I n, will have opposite phase polarity. Thus, if one corrects for the phase difference between the images and combines the images coherently, the visible artifacts will be noticeably reduced through signal cancellation. Here, we employ a subspace identification method 14 to identify an operator correct for the phase difference between I ' p and normalized cross-correlation matrix, conj{ } ' Ψ to ' I n. Specifically, we first compute the ( ) conj{ } Q= I oi I oi where o is an ' ' ' ' n p n n element-by-element matrix multiplication operator (Schur product), to calculate the phase difference between I ' p and ' n I. Then, the singular vectors, { uv, } r r, associated with the largest singular value of the matrix, Q, are identified through a singular value decomposition (SVD). A linear fit to the unwrapped phase of these individual vectors is then calculated. The slopes of these linear fits correlate to the k-space coordinate shift between ' k p and phase terms for in ' k n. The measured slopes of u r and v r are then used to form linear ' Ψ. The zero-order phase-difference term is then identified and included ' Ψ. Both corrections are then applied to one of the two images, aligning the phase between them. A distinct advantage of this SVD registration method is that phase unwrapping is performed on one-dimensional signals rather than a 2-D image, as both 73

88 u r and v r are vectors. A second advantage is that the projection of the phase difference data along rows and columns improves the algorithms' effectiveness in low SNR applications. Applying ' Ψ to the ' In image and then adding to I ' p produces an image free of Nyquist ghosts, due to the ghost cancellation effect described by the authors of PLACE 9. However, a disadvantage of the PLACE method is reduced temporal resolution. To overcome this limitation, we instead only employ this composite image data to calibrate parallel MR imaging reconstruction parameters. As we reported previously 15, eliminating ghosts in the calibration data has a dramatic effect on the quality of the parallel imaging reconstructions. The calculation of parallel imaging reconstruction coefficients is very sensitive to errors in calibration data, particularly in regards to phase information. Eliminating errors associated with Nyquist ghosts in self-referenced parallel MRI calibration data greatly enhances the applicability of these methods to EPI imaging. Thus, we use the Nyquist ghost free data, ( I + Ψ I ), to determine parallel ' ' ' p n imaging reconstruction coefficients for each frame. These coefficients are then used to reconstruct an image from the data acquired at time t, kt (). As in the real-time PAGE method 8, here we also separate the positive and negative readout data for each frame. In unaccelerated data sets, this separation creates two data sets, each effectively subsampled by a factor of 2 and acquired using the same readout gradient polarity. We then reconstruct the missing lines in each set using parallel imaging. One can use either GRAPPA-based 16 or SENSE-based 11 methods. To form the final image, we again estimate and employ an operator Ψ to correct for the zero- and first-order phase difference between the reconstructed images. This has the effect of noticeably reducing 74

89 residual pmri reconstruction artifacts, as again the images associated with each readout polarity, In( t ) and I p ( t ), will have similar artifacts of opposite phase. In the Results below, we employ GRAPPA to perform the pmri reconstruction steps. Thus, a rootsum-of-squares combination across the coil dimension of I () t + Ψ I () t is employed to form the final images. The exception to this is the example shown in Fig. 14, where a virtual body coil 17 calculation is employed to preserve phase. If SENSE is used in place of GRAPPA, the coil data combination step is unnecessary. The notable differences between our method and the real-time PAGE method described by Kim et. al 8 are shown in gray in Fig. 13, where we use PLACE to form composite images. Thus, instead of calibrating the pmri reconstruction coefficients on ' In and ' I p separately, we compute our pmri reconstruction coefficients on an image that is truly free of Nyquist ghosts. Similarly, rather than employing a root-sum-of-squares across the combined multi-coil data for both the positive and negative gradient readout images at time t, we construct the image for each frame through coherent addition to both maintain image phase and suppress residual artifacts. Implementation Details for Previous Methods: In the results below, the same measured data was reconstructed using multiple methods. These methods can be described relative to Fig. 13. For images reconstructed using the Ahn & Cho static method, the linear and constant phase correction parameters were estimated from a reference scan at the beginning of the acquisition with the phase encoding gradients set to zero (i.e. k y = 0 ). To simulate a dynamic Ahn & Cho correction from the same raw data, the linear and constant phase parameters were estimated from the EPI-GESTE images, rather than an extended EPI echo train. This change does not affect the comparison, as p n 75

90 the estimated values should be the same, and generating them from EPI-GESTE images is more reliable than from a limited number of k y = 0 phase-encode data lines. Real-time PAGE images were reconstructed in two stages. First, ' I p and ' I n were each used to estimate pmri reconstruction coefficients for the +G x and G x data at kt ( ), respectively. The reconstructed images were then combined using a root-sum-of-squares calculation across all 16 images 8 coils each of +G x and G x data. The PLACE images are equivalent to ( I ' +Ψ ' I ' ) in Fig. 13. To emphasize the loss in temporal resolution, p n PLACE images are referenced at time ( t 1/2). Data Acquisition The goal of the experiments presented in the Results is to highlight the advantages of our ghost elimination method compared to current standard methods of Nyquist ghost correction. In particular, in our experience with SNR-sensitive imaging methods such as ASL perfusion imaging, we have noticed subtle but significant shading and subtractionerror artifacts in ASL experiments when previous ghost correction methods were employed. [See Fig. 6, for example.] These experiments highlight these errors and demonstrate that EPI-GESTE corrects them. ASL Perfusion Imaging Experiment: Data for 124 image frames were acquired using an EPI sequence modified to acquire perfusion weighted images with a Q2TIPS- FAIR pulse sequence 18-20, using an 8-channel head coil on a GE EXCITE 1.5T scanner (v14m3) with a TwinSpeed gradient coil (GE Healthcare Inc., Milwaukee, WI). The imaging parameters were: image size = 128x128, TR/TE=2.5s/55.1ms, slice thickness = 8mm, FOV = 28mm x 28mm. Perfusion imaging is achieved by taking the difference 76

91 between two images: a label image where the magnetization of the inflowing blood is inverted or saturated, and a control image where the inflowing blood is fully relaxed. The different magnetic states are achieved by a pair of global and slice selective inversion pulses. When the control image is subtracted from the label image, the static tissue signal is removed. The remaining signal is proportional to the blood that has perfused into the imaged region. Control and label images are acquired alternating in time through the sequence. For this experiment, temporal encoding was introduced by shifting the k-space sampling grid by (1Δ k ) on every other control/label pair. 62 control and label pair differences (124 frames) were averaged together to produce final perfusion weighted image. Signal-to-Noise Measurement Experiment: To quantify the effect of our method on SNR, images of a structured water phantom were acquired using the alternating readout polarity trajectory. Twelve axial images were acquired using the acquisition parameters: TR/TE = 1sec/53msec; FOV=20cm; Slice thickness=5mm; image matrix size=128x128. Twelve double-oblique images were acquired using the acquisition parameters: TR/TE = 2sec/53msec; FOV=24cm; Slice thickness=5mm; image matrix size=128x128. SNR was measured globally by generating a mask representing the image region with signal at least 10% of the maximum pixel value for all images. The temporal mean and variance of the pixels both inside and outside this region were calculated across the 12 images in the series. 77

92 Ghosting Level Quantification Nyquist ghost levels were compared between methods by calculating the L 2 -norm of the pixel magnitude values in two 13-pixel-square boxes, one in a nearly constant region of the image, S s, and a second in a signal-void region with Nyquist ghosts, S g. The ghost to signal ratio (GSR) is then reported as a percentage of the ratio Sg S s. 2 2 Figure 14. An image-domain comparison of the pmri calibration data available in the method, with both magnitude (top) and phase (bottom) images shown. Image (a) shows significant ghosting in the image formed from interleaved data, caused by unintentional sampling inconsistencies between each of the two interleaved frames. Image (b) shows the ghosts are removed after applying a phase-alignment and addition operation. Note that in (a) and (b), pixels in the signal-void region are shown amplified by a factor of 5 to enhance ghost visibility. 78

93 RESULTS The results below illustrate the effectiveness of EPI-GESTE, which combines the Nyquist ghost suppression of PLACE within a real-time PAGE framework. Improvement in self-referenced parallel imaging calibration When temporal encoding is employed, Nyquist ghost reduction can often be achieved by interleaving two frames using data associated with one readout gradient polarity. When intentional (e.g. real-time FOV updates) or unintentional (e.g. scanner heating) changes in the acquisition occur between scans, however, interleaving alone may prove to be insufficient. An example of such an occurrence can be seen from the ASL perfusion experiment, where a significant unintentional change occurred between the acquisition of two frames. Fig. 14(a) shows the result of simple interleaving between two images frames using data acquired on the negative readout gradients. Sampling errors produced an inconsistency between the two frames, which produces significant visible ghosting in the interleaved image. In contrast, if one employs PLACE instead of simple interleaving, the residual ghosts will cancel to yield the image shown in Fig. 14(b). While excellent ghost elimination is achieved with PLACE, it comes at a cost of lowered temporal resolution due to data interleaving. To maintain the original temporal resolution in our method, this ghost-corrected interleaved data is used instead to calibrate the pmri reconstruction parameters. Removal of ghosts in this calibration data improves the quality of the subsequent image reconstructions that rely on these parameters. 79

94 Robustness to Scanner Instability The ASL perfusion imaging data was acquired on a system known to show increasing Nyquist ghost levels over the course of long image acquisition sessions. A possible reason for this is scanner field drift due to scanner heating 21. Figure 15. (a) A plot of drift in the estimated positive/negative readout gradient line shift correction coefficient as measured with EPI-GESTE over the image series. (b) A plot of the ghost-to-signal ratio as a function of data acquisition time, comparing five reconstruction methods. EPI-GESTE exhibits much lower GSR than Real-time PAGE or PLACE, some of which is attributed to the background noise suppression effect discussed in the text. Fig. 15(a) shows the effect of this field drift over the course of the 124 acquired ASL perfusion measurement images. Here, the value of the linear phase shift correction term, as calculated by the EPI-GESTE method, is plotted against image acquisition time. The plot shows a clear linear trend in this coefficient, indicating a slow drift in the relative position between the positive and negative EPI readout data of each temporal frame as the experiment progresses. Fig. 15(b) shows the measured ghost-to-signal ratio for each frame, measured at the locations shown in the boxes of Fig. 14. As expected, a static Ahn & Cho correction (Ahn & Cho static) shows a gradual increase in GSR levels over time, as the shift measurement was calculated only once at the beginning of the acquisition. Switching to a variable shift parameter (Ahn & Cho dynamic) removes this general trend. The real-time PAGE method 8 can reduce ghosts further, although the 80

95 correction tends to have wide variability indicated by the jump in GSR at frames 97, 99, 122, and 124. These spurious jumps are caused by significant ghosts in the interleaved data that corrupt the GRAPPA coefficient calibration at those time frames. This wide variability is absent in the PLACE method, because the ghosts appear equal in magnitude but with opposite phase, and are thus cancelled at each frame. This robustness in PLACE comes at a cost of decreased temporal resolution, however. 81

96 Figure 16. Image reconstructions to compare EPI Nyquist Ghost Correction methods: (a) Ahn & Cho static; (b) Ahn & Cho dynamic; (c) Real-time PAGE; (d) PLACE; and (e) EPI-GESTE. The squares in each image show the regions used to calculate the ghost-to-signal ratio shown in Fig. 3. The arrows highlight a ghost visible in previous methods that is not visible in PLACE or EPI- GESTE. Pixels within each drawn ellipse are shown on a standard gray scale, shown above. To enhance ghost visibility relative to the brain tissue signal, pixels outside of each ellipse are shown using a logarithmic gray scale shown on the upper right. The EPI-GESTE image (e) has notably reduced signal energy in the signal-void region outside the ellipse compared to other single frame methods (a-c). The cause of this effect is explained in the text and illustrated in Fig 8. 82

97 In contrast, our method (EPI-GESTE) shows a further reduction in GSR in Fig. 15(b). In addition to good ghost suppression, EPI-GESTE benefits from a background noise suppression effect that notably lowers the measured GSR. The source of this effect is described later in the SNR section of the text. Comparing for now the GSR time course variance for each of the spatial/temporal encoding methods, we measured: Real-time PAGE = ; PLACE = ; and EPI-GESTE = This demonstrates that EPI-GESTE has significantly lower ghost variability than the previous methods. Images from a particularly problematic point in time, t = 99, are shown in Fig. 16 for each reconstruction method. Note that to improve ghost visibility, pixels in the signalvoid region are shown on a logarithmic scale that is common across all images and spans the full gray-level scale. Specifically, a mask was determined from the EPI-GESTE image to cover all pixels greater than 1.6% of the maximum value. This mask was then filtered to remove any holes, using a 3x3 two-dimensional median filter. The base-10 logarithm of signal values outside of the mask was then computed. These signal-void values were then scaled such that the minimum value was zero and the maximum value in this region across all images was equal to the maximum gray level value within the ellipse. The signal-value to gray-level map for each region is shown in the upper left of Fig. 16. Here again, the ghost visibility in each image is consistent with the measured GSR levels shown in the previous plot, with the Ahn & Cho images showing much greater visible ghosting than the images reconstructed using spatial and/or temporal encoding. By combining the strengths of PLACE and real-time PAGE, EPI-GESTE is able to capture and correct errors that confound the real-time PAGE method at a higher 83

98 Figure 17. Images of discarded signal in (a) PLACE and (b) GESTE for a single time point. In each method, a coherent addition is performed to cancel residual artifact. In PLACE, this signal is dominated by the residual ghost signal present after interleaving. In GESTE, this signal is limited to artifacts from parallel imaging and field inhomogeneity effects. Significantly more signal is lost with PLACE, which can lead to unwanted variability in an image series. temporal resolution than PLACE alone. There is also a notable reduction in the background signal level of the signal-void region compared to the other images generated from a single frame of data (Fig. 16 a-c). The signal level in the background region of the GESTE image is, however, comparable to the two-frame PLACE image in Fig. 16d. This effect is discussed further in the SNR results below. Nyquist ghosts are visibly absent in both the PLACE and GESTE images of Fig. 16. However, Fig. 17 demonstrates that the GESTE image has higher fidelity to the measured data than PLACE. The figure shows two images of the signal that is lost during the coherent summation step in both PLACE, Fig. 17(a), and GESTE, Fig. 17(b). Specifically, Fig. 17(a) shows the coherent difference, I ' n Ψ I ' p, between the interleaved positive and negative gradient images that form the PLACE image. This illustrates the ghost artifact that is cancelled by PLACE, which for this frame is significant. In comparison, the coherent difference between the intermediate images that form the 84

99 GESTE image, I n Ψ I p, shows significantly less signal. The signal that is cancelled in the GESTE method is limited to parallel imaging artifacts and field inhomogeneity effects. This improves the robustness of the method for each time-point in an image series, by limiting the level of variability in the cancelled signal from time-point to timepoint. Fig. 18 shows the ASL perfusion images associated with the structural images shown previously, in Fig. 16. The perfusion image derived from the Ahn & Cho dynamic images, Fig. 18(a), shows reasonably good discrimination between the gray and white matter perfusion, consistent with the current standard in EPI ASL imaging. In contrast, the perfusion image associated with the real-time PAGE method, Fig. 18(b), shows the influence of the Nyquist ghost variations, with both ghost-related signal visible in the signal-void region, and an increase in overall signal intensity on the upper-left quadrant of the brain. Fig. 18(c) shows that with PLACE the ghosts are removed, but the brightening remains. With EPI-GESTE, both the ghosts and the brightening are absent. The presence or absence of this brightening is clearly seen in the images shown in Fig. 19, which show the arithmetic difference between the Ahn & Cho dynamic image (current standard practice) and the more modern methods. This difference image shows that the subtle brightening effect in the Real-time PAGE and PLACE perfusion images is due primarily to improperly corrected ghost signals. In contrast, Fig. 19(c) illustrates that the EPI-GESTE perfusion image in Fig. 18(d) shows no false perfusion brightening compared to the standard image in Fig. 18(a). 85

100 Figure 18. Perfusion images generated from the (a) Ahn-and-Cho dynamic, (b) real-time PAGE, (c) PLACE, and (d) EPI-GESTE methods. The perfusion data is shown with a normal gray scale inside the ellipse. Outside the ellipse, the signal-void region has been amplified by a factor of four, to enhance ghost visibility. The arrows point to regions affected by Nyquist ghosts, which appear as image brightening in the upper left brain quadrant and ghost artifacts in the signal-void region. 86

101 Figure 19. Difference images between the Ahn & Cho ASL perfusion image and the (a) Realtime PAGE, (b) PLACE, and (c) EPI-GESTE ASL perfusion images. The unusual perfusion brightening visible in the Real-time PAGE and PLACE images of Fig. 6 appear here as dark regions in the upper left quadrant of the brain region. The shading of these regions is consistent with incorrect ghost suppression. Effect on Noise / SNR Close inspection of the in-vivo EPI-GESTE images shown previously reveals that the noise level in the signal-void region appears to be comparable to the temporally averaged PLACE image and lower than the other ghost elimination methods. An evaluation of the water phantom image reconstructions was performed to better understand the source of this noise suppression. Fig. 20(a) shows an axial image of the water phantom used for the SNR experiments. Fig. 20(e) shows a double-oblique view of the same phantom. The remaining images in the figure show the signal variance across the temporal image series for each pixel, for the Ahn & Cho dynamic images (center left), the output of one pmri reconstruction channel, I( p ) from Fig. 13 (center right), and the EPI-GESTE images (far right). The Ahn & Cho images in Fig. 20(b,f) show moderate signal variance in the phantom region. In areas where ghosts appear, the signal variance is slightly higher than in the background. It is well known that SNR will decrease in accelerated parallel imaging applications, and this is consistent with the increase in signal variance shown in images (c) and (g). Notably, the noise in the signal-void region does not increase 87

102 Figure 20. Phantom (a,e) and signal variance images associated with (b,f) Ahn & Cho ghost correction, (c,g) the output of a single pmri reconstruction channel, I () t, and (d,h) the final EPI-GESTE reconstruction, I () t + Ψ I () t. The top row shows an axial view of the phantom. The bottom row shows a double oblique view. p n p significantly, as this area is unaffected by the coil sensitivities and thus fall outside the region of increased noise described by the g-factor 22, 23. When the final EPI-GESTE image is formed (d,h), however, two accelerated images contribute, I ' p and ' I n. This combination decreases the signal variance across the entire image, unguided by the coil sensitivity. This yields a return of the signal variance to unaccelerated levels in the signal region, while the noise in the signal-void region is also reduced to yield a reduction in signal variance in regions unaffected by the g-factor. This effect of noise suppression in the signal-void region is confirmed quantitatively, where the average signal variance level within the signal region and in the signal-void region was measured to be: axial images (b): , ; (c): , ; and (d): , ; and for double-oblique images (f): , ; (g): , ; and (h): , , respectively. That is, the signal variance in the 88

103 image region affected by coil sensitivity roughly doubles for each intermediate GRAPPA image, but then drops back to the original level when the two GRAPPA images, I () t and I ( t ), are combined. In contrast, at the GRAPPA image stage the mean variance in n the signal-void region drops slightly, from to for the axial images, due to the improved ghost suppression. When the two GRAPPA images are combined, the variance level in the signal-void region also drops in half as expected. We note that in comparable PLACE images, the noise variance is roughly one-half of the Ahn & Cho images due to averaging: measured at (0.0056, ) for the axial images; and (0.0036, ) for the double-oblique images. This implies that single image estimates of SNR, using a ratio between signal measured over regions of the phantom and a region in the signal-void, need to account for this effect. Pair-wise subtraction or multiple image measurements, e.g. 24, to estimate SNR with EPI-GESTE are preferable. Similarly, if one compares the GSR measurements of PLACE and EPI-GESTE in Fig. 14, one finds that the mean ratio over the time course p is 1.40 or roughly a factor of 2. This large difference is a direct result of the background noise suppression effect in EPI-GESTE. DISCUSSION We have demonstrated that our EPI-GESTE method, a fusion of previous temporal and spatial encoding methods to suppress Nyquist ghosts, achieves the goals we set at the outset. Specifically, through the use of temporal encoding, EPI-GESTE is selfreferenced. EPI-GESTE has low latency in that it requires data from only one previous image volume to calibrate the pmri reconstruction coefficients. We have also shown that 89

104 the method is more robust than previous methods to temporal or dynamic changes. In addition, we have shown the method is extremely consistent, and in perfusion imaging does not produce the subtraction errors that both the PLACE and the real-time PAGE methods may introduce. The method appears to be resistant to gradient imperfections and can be used successfully in oblique-plane EPI where such errors can be more pronounced 25. Most significantly, however, the achieved ghost suppression level is well below current standard expectations, with less than 1:5% GSR in the perfusion imaging example shown here. The large difference in GSR between EPI-GESTE and competing methods is attributable to the additional background signal suppression provided by the method. In some applications, such as ASL perfusion imaging, we have found this additional background suppression to be beneficial. We have demonstrated that either blip-shifting or readout gradient toggling can be used for temporal encoding. One advantage of using the alternating readout gradient method is that it maintains the same echo time for each acquisition. One potential concern is that the toggling of the readout gradients may produce slightly different eddy currents in the two acquired frames, which may lead to an inherent mismatch in the interleaved data. We briefly investigated this potential problem by measuring GSR levels after EPI-GESTE reconstruction with and without eddy current compensation provided by the manufacturer. We found no discernible difference in image quality or GSR levels. However, even if the difference in eddy currents was appreciable for certain imaging protocols, our method would still be robust to these differences. Each image in the final output series is reconstructed independently and the interleaved data is used only for 90

105 parallel imaging calibration. This calibration typically employs only low spatial frequency information. Thus, any eddy current effects would likely be inconsequential. One challenge of our method in real-time EPI applications is the repeated computation of the pmri reconstruction coefficients. This can be mitigated as described by Kim, et. al. 8, by performing the pmri calibration in a process separate from the image reconstruction pipeline and updating the pmri reconstruction parameters on an asneeded basis. ACKNOWLEDGEMENTS This work was supported in part by NIH U41 RR (PI:Jolesz), NIH R01 AA (PI:Daunais), the Functional Neuroimaging Lab at Brigham and Women's Hospital (Boston, MA), and the Center of Biomolecular Imaging at Wake Forest University Health Sciences (Winston-Salem, NC). The authors also extend special thanks to Dr. Santiago Aja Fern andez for discussion related to SNR. 91

106 REFERENCES 1. Chapman B, Turner R, Ordidge RJ, et al. Real-time movie imaging from a single cardiac cycle by NMR. Magn Reson Med. Sep 1987;5(3): Bruder H, Fischer H, Reinfelder HE, Schmitt F. Image reconstruction for echo planar imaging with nonequidistant k-space sampling. Magn Reson Med. Feb 1992;23(2): Ahn CB, Cho ZH. A new phase correction method in NMR imaging based on autocorrelation and histogram analysis. IEEE Trans Med Imaging. 1987;6(1): Jesmanowicz A, Wong E, Hyde JS. Phase correction for EPI using internal reference lines. Paper presented at: ISMRM 12th Annual Meeting, 1993; New York. 5. Yang QX, Posse S, Le Bihan D, Smith MB. Double-sampled echo-planar imaging at 3 tesla. J Magn Reson B. Nov 1996;113(2): Buonocore MH, Zhu DC. Image-based ghost correction for interleaved EPI. Magn Reson Med. Jan 2001;45(1): Chen NK, Wyrwicz AM. Removal of EPI Nyquist ghost artifacts with twodimensional phase correction. Magn Reson Med. Jun 2004;51(6): Kim YC, Nielsen JF, Nayak KS. Automatic correction of echo-planar imaging (EPI) ghosting artifacts in real-time interactive cardiac MRI using sensitivity encoding. J Magn Reson Imaging. Jan 2008;27(1): Xiang QS, Ye FQ. Correction for geometric distortion and N/2 ghosting in EPI by phase labeling for additional coordinate encoding (PLACE). Magn Reson Med. Apr 2007;57(4): Kellman P, McVeigh ER. Phased array ghost elimination. NMR Biomed. May 2006;19(3):

107 11. Pruessmann KP, Weiger M, Scheidegger MB, Boesiger P. SENSE: sensitivity encoding for fast MRI. Magn Reson Med. Nov 1999;42(5): Kellman P, Epstein FH, McVeigh ER. Adaptive sensitivity encoding incorporating temporal filtering (TSENSE). Magn Reson Med. May 2001;45(5): Pipe JG. Motion correction with PROPELLER MRI: application to head motion and free-breathing cardiac imaging. Magn Reson Med. Nov 1999;42(5): Hoge WS. A subspace identification extension to the phase correlation method. IEEE Trans Med Imaging. Feb 2003;22(2): Hoge WS, Tan H, Kraft RA. Improved self-referenced parallel MRI imaging in EPI by using UNFOLD to remove Nyquist ghosts. Paper presented at: ISMRM 17th Scientific Meetings, 2009; Honolulu, HI. 16. Griswold MA, Jakob PM, Heidemann RM, et al. Generalized autocalibrating partially parallel acquisitions (GRAPPA). Magn Reson Med. Jun 2002;47(6): Buehrer M, Boesiger P, Kozerke S. Virtual body coil calibration for phased-array imaging. Paper presented at: ISMRM 17th Scientific Meetings, 2009; Honolulu, HI. 18. Kim SG. Quantification of relative cerebral blood flow change by flow-sensitive alternating inversion recovery (FAIR) technique: application to functional mapping. Magn Reson Med. Sep 1995;34(3): Yang Y, Frank JA, Hou L, Ye FQ, McLaughlin AC, Duyn JH. Multislice imaging of quantitative cerebral perfusion with pulsed arterial spin labeling. Magn Reson Med. May 1998;39(5): Luh WM, Wong EC, Bandettini PA, Hyde JS. QUIPSS II with thin-slice TI1 periodic saturation: a method for improving accuracy of quantitative perfusion imaging using pulsed arterial spin labeling. Magn Reson Med. Jun 1999;41(6):

108 21. Brodsky EK, Samsonov AA, Block WF. Characterizing and correcting gradient errors in non-cartesian imaging: Are gradient errors linear time-invariant (LTI)? Magn Reson Med. Dec 2009;62(6): Larkman DJ, Nunes RG. Parallel magnetic resonance imaging. Phys Med Biol. Apr ;52(7):R Breuer FA, Kannengiesser SA, Blaimer M, Seiberlich N, Jakob PM, Griswold MA. General formulation for quantitative G-factor calculation in GRAPPA reconstructions. Magn Reson Med. Sep 2009;62(3): Sijbers J, den Dekker AJ, Van Audekerke J, Verhoye M, Van Dyck D. Estimation of the noise in magnitude MR images. Magn Reson Imaging. 1998;16(1): Reeder SB, Atalar E, Faranesh AZ, McVeigh ER. Referenceless interleaved echoplanar imaging. Magn Reson Med. Jan 1999;41(1):

109 CHAPTER V PSEUDO-CONTINOUSL ARTERIAL SPIN LABELING 95

110 ABSTRACT Arterial spin labeling (ASL) is a non-invasive MRI technique for measuring cerebral blood flow by magnetically labeling the inflowing arterial blood, which is usually accomplished using either intermittent (PASL) or continuous pulses (CASL). An improved labeling method, known as pseudo-continuous ASL (PCASL), has been reported to combine the advantages from both PASL and CASL to obtain higher perfusion signal-to-noise ratio (SNR) than PASL and improved labeling efficiency than CASL. In this study, balanced PCASL was implemented with 3D GRASE PROPELLER readout sequence and background suppression pulses. The perfusion measurements were performed for in human volunteers and non-human primates (NHPs) on 1.5T and 3T scanner, respectively. Unexpectedly, no significant improvement on perfusion SNR was observed in human studies compared to PASL, which requires further investigation. In contrast, the perfusion sensitivity and image quality were much improved for NHPs. The results suggest PCASL is a feasible technique to measure brain perfusion information in NHPs on human scanners with reasonable spatial resolution, thereby setting the stage for developing a non-invasive technique for quantitative CBF measurement in NHP. 96

111 INTRODUCTION Over the last decade, arterial spin label (ASL) has rapidly become a popular magnetic resonance imaging (MRI) technique for cerebral blood flow (CBF) measurement in both clinical and research environments. Unlike the conventional techniques that rely on an external contrast agent (e.g. Dynamic Susceptibility Contrast MRI, Positron Emission Tomography), ASL uses the water in the arterial blood as an endogenous tracer to measure perfusion, and therefore is completely non-invasive and highly repeatable. While widely used in human, ASL technique has just begun to be applied in non-human primate (NHP) studies in recent years 1, 2. NHP provides an important animal model recapitulating many human diseases due to its overall similarities to human. However, it is difficult to use conventional ASL techniques to measure brain perfusion in NHPs because of its small anatomical size. Meanwhile, NHPs are often too big to fit into an animal scanner that is often equipped with a stronger magnetic field. Hence, NHP studies usually take place in human scanner. Under those limitations, a suitable ASL method and protocol for NHP perfusion imaging is still to be determined. There are two main classes of ASL, pulsed ASL (PASL) and continuous ASL (CASL) that are widely adapted (see Chapter I). PASL has a high labeling efficiency and is insensitive to variation in blood flow velocity. However, it has poor perfusion signalto-noise ratio (SNR). Compared to PASL, CASL has a higher perfusion SNR, but the availability CASL is limited due to the hardware requirement for continuous radio frequency (RF) transmission. Furthermore, the labeling efficiency of CASL can be affected by flow velocity and is generally lower than PASL 3. In recent development, a new form of ASL, referred to as pseudo-continuous ASL (PCASL), is introduced to 97

112 combine CASL s high SNR with PASL s superior labeling efficiency. In PCASL, the continuous RF waveform in CASL is replaced with a train of discrete RF pulses to mimic the flow-driven adiabatic inversions 3-5. As a result, PCASL has been shown to achieve a 50% increase in SNR compared to PASL and a higher labeling efficiency than CASL 3, 6. Those advantages make PCASL a desirable alternative for ASL studies. Combining PCASL with advanced 3D readout sequence, the ASL technique can be optimized to minimize scan time and improve image quality. The objective of the work reported in this Chapter is to explore the optimal PCASL parameters, in addition to evaluating the performance and applicability of PCASL for NHP studies. PCASL was implemented with 3D GRASE PROPELLER readout sequence (see Chapter III) and background suppression 7 to maximize perfusion sensitivity. The perfusion SNR of PCASL was compared to PASL at 1.5T and 3T scanners for both human and NHP studies, respectively. Technical issues associated with PCASL implementation are discussed. 98

113 Figure 21. Balanced PCASL tagging scheme for two successive RF pulses (a) and the entire pulse sequence diagram (b). ISat and OSat stand for in-plane saturation and outer plane saturation, respectively. Figure (a) is taken from literature 3. MATERIALS AND METHODS PCASL Implementation Balanced PCASL 3, 8 was developed due to the ease of implementation. The pulse train for two successive RF pulses of balanced PCASL is shown in Figure 21a. The 99

114 polarity of the RF pulses is constant during the labeling state and opposite during the control state. The gradient waveforms are identical between control and label states, however, with a non-zero net gradient movement per cycle (i.e. between any two consecutive RF pulses). The labeling efficiency of PCASL is dependent upon two factors: the residual gradient moment, denoted as η, and the flip angle of the RF pulse, denoted as α. The residual gradient moment is defined as the ratio between the net difference gradient area and the positive gradient area, or + A A η = 100% + A where A + and A are areas under positive and negative gradient periods, respectively. η is adjusted by changing the amplitude of the negative gradient. In our implementation, η was set to 10% and α was set to 20 to maximize perfusion sensitivity 3, 4. Hanning window shaped RF pulses were employed with a duration of 500 μs. The body coil was used for transmission and an eight-channel phased array coil was used for data collection. Background Suppression and Image Acquisition Very selective suppression (VSS) pulses 9 were employed at the beginning of the sequence to saturate the imaging slab prior to PCASL labeling (pre-saturation), followed by the PCASL labeling pulse train. Two inversion pulses (non-selective for human and slice-selective for NHP) and outer plane saturation pulses were played out after the PCASL labeling sequence to null the static tissue signal before image acquisition. The purpose of the background suppression pulses was to reduce imaging artifacts and dynamic range requirements 4, 7. 3D GRASE PROPELLER (3DGP) was employed as the image acquisition sequence. The full pulse diagram is shown in Figure 21b. 100

115 MR Experiments Perfusion imaging experiments were conducted in both human and NHP. The imaging environment and parameters are described below. Two healthy male volunteers (age 28 and 46) were imaged on a GE 1.5T Signa TwinSpeed scanner (GE Healthcare System, Milwakee, MI). The human subjects were recruited under the guideline of the Institutional Review Board, and informed written consent was received from each subject. The labeling duration of PCASL was set to 1.5 seconds with a post labeling delay time of 1 second. A time of flight (TOF) scan was prescribed to determine the labeling location. The purpose was to ensure the labeling plane was perpendicular to the main feeding arteries of the brain (e.g. internal carotid and vertebral arteries). Perfusion images were acquired with 3DGP. For subject one, 8 bricks of matrix size were acquired per image with a field of view (FOV) of mm, resulting in a nominal voxel size of mm 3. For subject two, the imaging parameters were slightly different the brick matrix size is with a FOV of mm yield in a nominal voxel size of mm 3. Partial Fourier encoding was used in both subjects to reduce the number of actual partitions to 13. The total scan time was about 4 minutes. NHP experiments took place on a GE 3T Signa Excite scanner. An adult, male Rhesus monkey (14 years old, 9.5kg) was used in the perfusion study. The animal was anesthetized and intubated, and placed in a supine position inside the scanner bore. A custom designed flexible 8-channel head coil was used for reception. End-tidal CO 2, O 2 saturation, heart rate, respiration rate, and temperature were monitored continuously and maintained within normal physiological range during the entire experiment. Similar to 101

116 human imaging, a TOF scan was used to determine the labeling location. The labeling duration of PCASL was set to 2.1 seconds with a post labeling delay time of 1 second. A total of 16 slices were acquired in two-shot to reduce T 2 -induced through-plane blurring in acquisition. Other imaging parameters include: the number of 3DGP bricks = 8, blade size = 20 96, FOV = 192 mm, nominal voxel size = mm 3. The total scan time was about 10 minutes. PASL images, acquired with Q2TIP-FAIR 10, 11, were used in comparison for assessing the PCASL image quality. Due to the time constraint, PASL images in one human study were acquired from a different scan session with the same subject. In the NHP study, PASL images used for comparison were obtained from a different animal (rhesus monkey, 10 years old, 7.8 kg). While no quantitative comparison can be made between images acquired in different animals, the purpose here is to provide a qualitative gross assessment on perfusion sensitivity and image quality. The same image acquisition (3DGP) was used for PASL with similar parameters and scan time unless otherwise noted. The imaging pulse sequence used in those studies was developed internally and all imaging data was reconstructed offline using MATLAB (Natick, MA). Perfusion SNR was calculated based on region-of-interest (ROI). It was pointed out in Chapter IV that the GESTE 12 algorithm changed the distribution of the background noise in the final reconstructed image. Hence, ROI-based SNR analysis was not recommended for images processed by GESTE. However, in this situation where both PASL and PCASL images were acquired using the same readout sequence and processed with identical reconstruction algorithm, it is reasonable to assume the noise distributions in the background are identical for both sets of images. While the ROI-SNR 102

117 analysis cannot reveal the true SNR, it can however, be used to determine which method has higher perfusion sensitivity between PCASL and PASL as a relative comparison method. The ROI was manually drawn within similar anatomical regions (e.g. gray matter) on both sets of images. Multiple ROIs were used across multiple slices to provide a final averaged SNR for each method. RESULTS Figure 22 shows the final perfusion weighted image from human subject one. The PASL and PCASL scans were acquired during the same scan session. ROI-SNRs for both methods were calculated: SNR PASL = 22.4 and SNR PCASL = While both sets of perfusion weighted images demonstrated high quality visually, the expected SNR gain from PCASL was not obvious. Perfusion weighted images from human subject two are shown in Figure 23. PASL image was acquired from a previous study prior to the PCASL experiment. The number of bricks used in the PASL acquisition was doubled compared to the PCASL acquisition, which was taken into account during SNR analysis. The measured ROI-SNR was 20.9 for PASL and 17.5 for PCASL. Accounting for the difference in the number of 3DGP bricks used, the PASL SNR was rescaled to = PCASL showed an 18% increase in SNR from PASL, however, was still much less than the expected 50%. 103

118 Figure 22. Comparison of PCASL and PASL images acquired from human subject one. Both sets of images are shown with the same window/level. The receiver gains for both methods were the same during acquisition. The PCASL and PASL datasets were acquired in the same scan session with identical imaging parameters and acquisition sequence. 104

119 Figure 23. Comparison of PCASL and PASL images acquired from human subject two. Both sets of images are shown with the same window/level. The receiver gains for both methods are taken into account to scale the image properly. PCASL and PASL datasets were acquired during different imaging sessions. Same image acquisition sequence was used with similar imaging parameters. 105

120 Figure 24. Comparison of PCASL and PASL images acquired from the NHP study. Images were scaled accordingly. Note the PASL scan is obtained from a different animal. The number of imaging slices and the nominal slice thickness are different in the two datasets. The nominal inplane resolution is the same. 106

121 In the NHP experiment, the PCASL images were acquired with two-shot while the PASL images were acquired with a single shot. However, the number of repetitions used for signal averaging was the same for both datasets. The slice thickness of the PASL images was 4 mm, resulting in an approximately 33% larger voxel size than the PCASL images (slice thickness = 3 mm). Even with a higher resolution along the slice encoding direction, the PCASL images demonstrated higher perfusion SNR compared to the PASL images, as illustrated in Figure 24. Better anatomical detail such as the subcortical structures was clearly revealed in the PCASL images. DISCUSSION Based on the labeling schemes, current ASL techniques can be roughly divided into two categories. PASL, which uses a short RF pulse of a few milliseconds to tag the arterial blood through a thick labeling plane, has high tagging efficiency but low perfusion SNR. CASL, which employs a long and continuous RF pulse for a few seconds to tag the arterial blood through a thin labeling plane, has higher perfusion SNR than PASL but lower tagging efficiency. The SNR advantage of CASL is further offset by the hardware needs of continuous RF transmission that is not available on common scanners. In recent development, a new tagging method, referred to as pseudo-continuous ASL (PCASL), was developed with the intention to combine the merits of high SNR from CASL without the hardware requirement, and the high tagging efficiency from PASL. The spin labeling is achieved in PCASL by a train of repeating RF pulses with a gradient waveform of non-zeroth moment applied between each pair of the RF pulses. The control and label state is achieved by toggling the polarity of every other RF pulse in the pulse 107

122 train. The PCASL implementation in our studies is referred to as the balanced PCASL due to fact that the gradient waveform in the tagging sequence is the same between control and label state. Other implementations, such as unbalanced PCASL 3, 5, have been shown to achieve similar performances, which is beyond the scope of this chapter. The remaining chapter will focus on the balanced PCASL, although some of those points in the discussion also apply to other versions of PCASL. One critical factor to ensure the high tagging efficiency is that the phase between the two successive RF pulses must follow the phase evolution of the flowing spins 13. In balanced PCASL, the phase accumulation caused by the non-zero residual gradient moment must be taken into account in each corresponding RF pulse. Not doing so can result in large and unpredictable variations in the temporal series of perfusion weighted images, leading to significantly decreased tagging efficiency, lost perfusion SNR and erroneous CBF quantification. There are other sources of errors such as imperfect gradient systems and off-resonance fields that can reduce the inversion efficiency, which is further discussed in Chapter VII. PCASL is employed in combination with a 3D acquisition sequence to optimize perfusion sensitivity. One advantage of the 3D acquisition sequence is that the background suppression can easily be incorporated into the pulse sequence since there is only a single excitation pulse. Two adiabatic inversion pulses are played out after the PCASL tagging to invert the signal in the entire sample. Based on the PCASL tagging duration and the post-labeling delay time, the timing of the two inversion pulses is chosen to reduce the signal from the static brain tissue during image acquisition. In our experience, poor tissue suppression can cause significant subtraction errors between the 108

123 control and label images. About 95% of tissue suppression is ideal for ASL perfusion imaging 4. The two inversion pulses are played out at 21 ms and 419 ms after the PCASL tagging pulses for human experiments and 20 ms and 465 ms for the NHP experiment. Note that slice-selective inversion pulses for background suppression are used for NHP imaging. This is to ensure the background suppression pulses will not disturb the spin state of the freshly delivered arterial blood prior to its arrival in the tagging plane. The perfusion sensitivity was much improved with PCASL for the NHP experiment. This was an encouraging result which matched the literature findings 14, 15. It was shown that PCASL has the potential to be a viable solution for NHP perfusion imaging. Quantitative comparison between PASL and PCASL for NHP imaging is needed in the future for validation. No significant SNR improvement was observed in the human study on the 1.5T scanner which was unexpected. Since the sequence implementations for PASL and PCASL were fundamentally different, the amount of background suppression achieved in both methods was different which may be one of the reasons for reduced perfusion SNR for PCASL. Incorrect tagging locations (e.g. the labeling planes and labeling arteries are parallel with each other) may also lead to reduced labeling efficiency. Furthermore, the labeling efficiency can be compromised due to phase errors at the labeling location as a result of off-resonance fields, eddy current and gradient imperfections (see Chapter VII). Further investigation is warranted to determine the cause of the reduced PCASL performance in the human study. 109

124 CONCLUSION The work shown in this Chapter presented our initial experience with PCASL on both 1.5T and 3T systems in humans and NHP. The apparent lack of improvement in perfusion SNR in the human studies was unexpected and requires future work to explore optimal labeling parameters. In the animal study, PCASL demonstrated higher image quality than PASL, thus showing the potential to be a suitable tagging scheme for NHP perfusion imaging. While the NHP results were promising, the imaging resolution was still relatively low and the imaging time required for signal averaging was long. Further work will continue to focus on improving the robustness and perfusion sensitivity of the PCASL implementation. 110

125 REFERENCE 1. Zhang X, Nagaoka T, Auerbach EJ, et al. Quantitative basal CBF and CBF fmri of rhesus monkeys using three-coil continuous arterial spin labeling. Neuroimage. Feb ;34(3): Zappe AC, Reichold J, Burger C, et al. Quantification of cerebral blood flow in nonhuman primates using arterial spin labeling and a two-compartment model. Magn Reson Imaging. Jul 2007;25(6): Wu WC, Fernandez-Seara M, Detre JA, Wehrli FW, Wang J. A theoretical and experimental investigation of the tagging efficiency of pseudocontinuous arterial spin labeling. Magn Reson Med. Nov 2007;58(5): Xu G, Rowley HA, Wu G, et al. Reliability and precision of pseudo-continuous arterial spin labeling perfusion MRI on 3.0 T and comparison with (15)O-water PET in elderly subjects at risk for Alzheimer's disease. NMR Biomed. Dec ;23(3): Dai W, Garcia D, de Bazelaire C, Alsop DC. Continuous flow-driven inversion for arterial spin labeling using pulsed radio frequency and gradient fields. Magn Reson Med. Dec 2008;60(6): Fernandez-Seara MA, Edlow BL, Hoang A, Wang J, Feinberg DA, Detre JA. Minimizing acquisition time of arterial spin labeling at 3T. Magn Reson Med. Jun 2008;59(6): Ye FQ, Frank JA, Weinberger DR, McLaughlin AC. Noise reduction in 3D perfusion imaging by attenuating the static signal in arterial spin tagging (ASSIST). Magn Reson Med. Jul 2000;44(1): Wong EC. Vessel-encoded arterial spin-labeling using pseudocontinuous tagging. Magn Reson Med. Dec 2007;58(6): Tran TK, Vigneron DB, Sailasuta N, et al. Very selective suppression pulses for clinical MRSI studies of brain and prostate cancer. Magn Reson Med. Jan 2000;43(1):

126 10. Kim SG. Quantification of relative cerebral blood flow change by flow-sensitive alternating inversion recovery (FAIR) technique: application to functional mapping. Magn Reson Med. Sep 1995;34(3): Luh WM, Wong EC, Bandettini PA, Hyde JS. QUIPSS II with thin-slice TI1 periodic saturation: a method for improving accuracy of quantitative perfusion imaging using pulsed arterial spin labeling. Magn Reson Med. Jun 1999;41(6): Hoge WS, Tan H, Kraft RA. Robust EPI Nyquist ghost elimination via spatial and temporal encoding. Magn Reson Med. Dec 2010;64(6): Jung Y, Wong EC, Liu TT. Multiphase pseudocontinuous arterial spin labeling (MP-PCASL) for robust quantification of cerebral blood flow. Magn Reson Med. Sep 2010;64(3): Wey HY, Li J, Jones L, et al. Quantitative CBF MRI of Anesthetized Baboon using Pseudo-continuous ASL. Proceedings of the 17th Annual Meeting of ISMRM Wey HY, Li J, Wang J, Park SH, Duong TQ. Blood-flow MRI of Non-human Primate (Baboon) Retina. Proceedings of the 17th Annual Meeting of ISMRM

127 CHAPTER VI PERFUSION PHANTOM VALIDATION USING ARTERIAL SPIN LABELING MRI 113

128 ABSTRACT The lack of a true standard in arterial spin labeling (ASL) imaging causes incompetent comparisons and systematic biases among various ASL techniques. A quantifiable perfusion phantom will fill in the gap for comparing and validating ASL techniques. An MR-compatible, flow-based perfusion phantom was developed by researchers at Virginia Polytechnic Institute and State University and validated initially using thermodynamics. However, the ASL validation experiment failed due to a long transit time as a result of the phantom dimensions and a limited flow rate in the current setup. Redesign of the perfusion phantom and improvements in the flow system are required for further investigation. 114

129 INTRODUCTION Arterial spin labeling (ASL) is a non-invasive magnetic resonance imaging (MRI) technique that has been widely used for quantitative measurement of perfusion. ASL uses water in the arterial blood as an endogenous tracer to investigate perfusion in by acquiring two separate images: a label image where the signal from the inflowing blood has been inverted or saturated; and a control image where the signal from the inflowing blood is unchanged. By subtracting the label image from the control image, the difference signal is proportional to the blood that has flown into the imaging region. Although many versions of the ASL technique exist, only in-vivo data are used for method comparison and validation. While largely accepted by the ASL community, it is unfitting to use in-vivo data due to the large number of uncontrollable variables in in-vivo experiments. Factors such as psychological conditions (e.g. blood flow rate), subject motion and substances intake (e.g. caffeine, cigarette) have significant impact on the quality and quantification of perfusion images. It is a difficult task to keep all these circumstances identical among subjects and experiments, which introduces biases and errors in perfusion quantification. This makes validation of in-vivo results less accurate, especially when comparing across studies with different ASL techniques. A perfusion phantom would be an ideal solution for this problem. It has several advantages over invivo experiments. First, there will not be any motion artifacts and partial volume effects since the phantom will be stationary and only one material is present. Second, a phantom has no physiological noise due to cardiac and respiratory motion. In addition, the perfusion quantification from a phantom is expected to be highly reproducible, favorable 115

130 for eliminating quantification biases among different existing methods, and for new method development and validation. In this study, a flow-based perfusion phantom was designed by our collaborators at Virginia Polytechnic Institute and State University to model the cerebral perfusion process. The perfusion phantom was designed to mimic realistic psychological values and was validated with thermodynamics. The goal of this study is to validate the perfusion phantom using ASL method and to evaluate the applicability of using such a phantom for the future development of MR perfusion methods. MATERIALS AND METHODS ASL Sequence Clinical perfusion images were acquired at Wake Forest University Baptist Medical Center as a routine clinical practice 1. The quantitative perfusion maps are measured with QUantitative Imaging of Perfusion using a Single Subtraction with Thin Slice TI1 Periodic Saturation (QUIPSS II TIPS a.k.a. Q2TIPS) 2 with a Flow-sensitive Alternating Inversion Recovery (FAIR) 3. In our implementation of Q2TIPS-FAIR, saturation pulses are Very Selective Suppression (VSS) radio frequency pulses 4, which are applied every 25 ms between 800 ms (TI1) and 1200ms (TI1s). The VSS pulses saturate a 2 cm slab of tissue with a 1 cm gap between the saturation slab and the first imaging slice. Our sequence uses C-Shaped Frequency Offset Corrected Inversion (FOCI) pulse (β=1361, μ=6) to reduce slice imperfections and improve sensitivity 5. 2D EPI is used as the image acquisition sequence. Other imaging parameters are as follows: TE = 28ms, TI1 = 800ms, TI1s = 1200ms, TI = 2000 ms, TR = 3000 ms, receiver bandwidth 116

131 62.5 khz, flip angle 90 degrees, FOV 24 cm, an acquisition matrix 64 x 64 with 5 mm slice thickness. Perfusion Phantom Design The primary component of the perfusion phantom consists of a cylindrical tube with two connectors located on the opposite ends of the tube where fluid (water) can be pumped from one side and exit from the other. The key element distinguishing the perfusion phantom from a pure flow phantom 6 is the perfusion tissue, a cloth that separates the cylindrical tube into two chambers: a mixing chamber and a perfusion chamber. This design is to allow the bolus of freshly delivered (labeled) fluid to fully mix with the existing fluid in the mixing chamber, mimicking the cerebral perfusion process. The mixed fluid then moves through the perfusion tissue in the perfusion chamber where the perfusion signal will be measured. The schematic of the perfusion phantom is illustrated in Figure 25a. The inlet and outlet tubes (terminal connectors) are connected to the flow pump to generate a constant flow in the phantom. When the inflowing fluid enters the mixing chamber through a considerably smaller diameter, a jet (non-directional flow) is formed as a result of the jet impinging the perfusion tissue. This in theory allows the new fluid to fully mix with the existing fluid. The mixed fluid then moves through the perfusion tissue into the perfusion chamber with uniform flow where the perfusion measurement will take place. The dimensions of the perfusion phantom are calculated to meet two design specifications: 1) the phantom can be placed horizontally inside the scanner bore; 2) Time of perfusion (for the tagging fluid to mix with existing fluid and move pass through the perfusion tissue) is 117

132 2 seconds in order to match the ASL protocol. The picture of the actual phantom is shown in Figure 25b. The area ratio between the phantom cross-section and inlet tube cross-section is 50:1 to ensure turbulent flow for complete mixture of the inflowing fluid. Figure 25. The schematics of the perfusion phantom design (a) and a picture of the actual phantom (b). The red arrow bands in (a) illustrate the flow pattern in the mixing and perfusion chambers. 118

133 Perfusion Model As suggested in the existing literature 7, perfusion is quantified as the ratio of the flow rate of the system and the passage volume, that is Flow Rate of System (Q) Perfusion(P) = Volume of Mixing Chamber (V) which can be expressed as Q V * A V P = = = V L * A L PC IT PC MC IT MC where V PC is the fluid velocity though the perfusion chamber, A IT is the area of the crosssection and L MC is the length of the mixing chamber (Figure 25a). Perfusion Validation Using Thermodynamics The initial results on the perfusion phantom were obtained with the idea of heat transferring to mimic the ASL technique. The thermal event was created by injecting new fluid of different temperature to the existing fluid in the mixing chamber of the phantom. The fluid temperature was recorded immediately after the perfusion tissue with thermocouples. The positions of the thermocouples were illustrated in Figure 26. Consistency of the thermocouple temperatures was also checked during the time of perfusion to ensure that there was complete mixing in the mixing chamber and uniform flow in the perfusion chamber. The thermal event used to model the perfusion was based on the first law of thermodynamics, shown in the following equation P t ( T T ) = ln t ( T T ) 0 119

134 where P t was perfusion at time t, T was measured temperature at time t, T 0 was the initial temperature (temperature of the fluid in the mixing chamber before the entry of the new fluid) and T was the equilibrium temperature. The averaged perfusion value could be found by computing the slope of the linearly fitted P t values over time. This heat transferring perfusion model assumes well-mixed chamber and negligible heat loss through perfusion volume. Figure 26. The locations of the four thermocouples on the perfusion tissue for the thermodynamic validation. Flow Measurement and Validation In order to determine the flow rate achievable with the existing flow pump system, phase contrast MRI was used to measure the flow rate within the system. The perfusion phantom was connected to the flow pump and placed horizontally inside the scanner. A neurovascular coil was used for data collection. Steady flow was established before the measurements were conducted. The imaging location of the phase contrast measurement was placed at the inlet tube. The imaging parameters included: TR = 500 ms, flip angle = 120

135 20, field of view = 200 mm, slice thickness = 5 mm, velocity encoding = 75 cm/sec. The flow rate value was determined by taking the mean value of the region of interests (ROI) manually drawn inside the inlet tube. The magnitude image was used to assist determining of the ROI boundary. Flow rate results obtained from the phase contrast MRI was validated using the conventional timed volume collection measurement. The outflow fluid was collected in units of 500 ml and the duration time was recorded. The measurement was repeated and the average flow is calculated by Volume 500 ml flow = = Mean Collection Time n 1 T n i= 1 i RESULTS Perfusion Results Using Thermodynamics Fluid entering the mixing chamber was switched from hot to cold. Temperature data was recorded by the thermal couple placed on the perfusion tissue. Three experiments with variable flowrates (26.6, 19.2 and 16.6 ml/s) were conducted. The phantom was positioned vertically where inflowing fluid enters the mixing chamber from the top. The results from the three experiments are shown in Table 5. Table 5. Perfusion results from thermodynamic validation Experiment Flowrate System Calculated Perfusion Validated Perfusion Index [ml/s] [ml/100g/min] [ml/100g/min] % Error

136 The differences between calculated and validated perfusion values were found to be small. This method of thermodynamics for testing the perfusion phantom has seen to be accurate and repeatable. Phase Contrast Results and Timed Volume Collection Validation The perfusion phantom was placed horizontally inside the scanner bore for the phase contrast measurement. The flow direction was from the superior end to the inferior end. The ROI was drawn in the inlet tube from the high resolution 3DSPGR magnitude image with an area of 0.58 cm 2. The mean measured flow was ml/sec. The phase contrast results were validated using timed volume collection. Five measurements were made to collect 500 cm 3 : 43, 44.7, 43.9, 45.1 and 46.4 sec. The mean flow measured was ml/sec, which matched the phase contrast result closely. Due to hardware limitations, an average 11.5 ml/sec was the maximum flow rate achievable within the current setup. Initial MR Experiment Result The perfusion phantom was placed in the same position (horizontally) as the phase contrast measurement inside the scanner bore. Steady flow rate (~12 ml/sec) was established prior to imaging. Imaging slices were prescribed immediately after the perfusion tissue in the perfusion chamber. 60 control/label pairs were acquired just under 7 minutes for signal averaging. The results, however, were inconclusive as there was no recognizable perfusion signal in the reconstructed perfusion weighted images. One highly 122

137 likely cause is the extended transit time during which the signal from the tagged bolus has completely decayed away before reaching the imaging plane. To verify this hypothesis, a flow experiment with red dye injection was performed to estimate the transit time of the tagged bolus. Figure 27. Flow experiment with red dye. The phantom is positioned in horizontal (a) and vertical position (b). Note the flow pattern across the perfusion tissue is much more uniform in the vertical position than the vertical position. The screenshot is captured 30 seconds after the initial dye enters the mixing chamber for the horizontal position (a) and 10 seconds for the vertical position (b). In the current system where the maximum achievable flow rate was about 12 ml/sec, the total tagged volume was 12 ml/sec 0.7 sec (tagging duration) = 8.4 ml. Hence, approximately 8.5 milliliter of dyed water was injected into the flow system to mimic the tagged bolus. The transit time was estimated from the time duration when the dyed bolus first entered the mixing chamber until its initial passage through the perfusion tissue. A video camera was used to capture the process and record the transit time. With the phantom positioned horizontally, the dyed fluid initially passed the perfusion tissue seconds after entering the mixing chamber. The delivery of the dye also presented an unusual pattern, illustrated in Figure 27a. We do not yet fully understand the cause of the non-uniform passage across the perfusion tissue, which may be the result of slow 123

138 flowrate and the density of the dyed fluid. The experiment was repeated with the phantom positioned vertically. The first group of dyed fluid crossed the perfusion tissue in 5-6 seconds. Although a significant improvement over the horizontal position, the transit time was still too long compared to T 1 of water. In addition, the phantom would not be able to fit inside the scanner bore vertically in its current dimension. However, one important observation was that the flow pattern across the perfusion tissue was much more uniform (Figure 3b). At last, this experiment revealed the long transit time was the primary reason that no signal was detected by the MR experiment. DISCUSSION The existing phantom in the current setup is not applicable for MR ASL validation due to a long transit time, which exceeds the T 1 of the fluid (water) resulting in complete decay of the tagged bolus signal prior to imaging. The long transit time is a consequence of limited flow rate, mixing chamber volume and the position of the phantom. To improve the MR experiment, the following modifications should be considered. First, higher flow rate in the system is needed. However, it should be noted that the flow rate adjustment should not surpass the threshold that causes too much turbulence in the mixing chamber leading to non-uniform flow across the perfusion tissue in the imaging plane. Second, the mixing chamber volume can be reduced to both shorten the bolus travelling distance and increase the fluid velocity inside the phantom body. Cautions should be taken that the width of the mixing chamber needs to be long enough to prevent the impinging jet flow from the inlet tube pushing through the perfusion tissue 124

139 within the given flow rate range. Third, the phantom needs to be placed vertically inside the scanner to ensure more uniform flow across the perfusion tissue. Relative perfusion proportion (RPP), defined here as the proportion of the tagged bolus volume and the total imaging volume, should also be considered. Assuming a complete delivery of the tagged bolus, a perfect imaging slice profile and ignoring relaxation effects, the relative perfusion sensitivity can be computed as Volume of Tagged Bolus RPP = 100% Imaging Volume Tagging Duration Flow Rate = 100% Phantom Cross Area Imaging Slice Thickness This information may be used to guide imaging prescription and to determine the transit Figure 28. Relative perfusion proportion is plotted as a function of the imaging slice thickness based on the existing phantom and experiment design. 125

140 efficiency (i.e. the actual amount of the tagged bolus delivered to the imaging plane at the time of imaging). For demonstration purpose, the RPP of the current system can be plotted as a function of the slice thickness, shown in Figure 28. This reveals, for instance, if 5 mm slice thickness is prescribed, the perfusion signal will be less than 2% of total signal this information can help determine the optimal imaging parameters. Furthermore, by comparing the RPP to the ratio of the mean magnitude signal between the perfusion weighted image and the control image, one can estimate the actual amount of detected bolus signal to estimate the transit efficiency of the system to optimize parameters such as flow rate and transit delay time (TI). CONCLUSION The current perfusion phantom design and the flow system setup are inadequate for MR testing. Modifications such as flow rate and phantom dimensions are needed prior to further MR testing. Transit time of the tagged bolus is a key factor to the success of the phantom system and should be carefully modeled. Further investigation of this project is required. ACKKNOWLEDGEMENT The author thanks our collaborators at Virginia Polytechnic Institute and State University, Dr. Thomas Diller and his students Elias Derke, Everett Irby, Charles Jacobson and Jesse Bernardo for building and validating the perfusion phantom using thermodynamics. 126

141 REFERENCE 1. Maldjian JA, Laurienti PJ, Burdette JH, Kraft RA. Clinical implementation of spin-tag perfusion magnetic resonance imaging. J Comput Assist Tomogr. May- Jun 2008;32(3): Luh WM, Wong EC, Bandettini PA, Hyde JS. QUIPSS II with thin-slice TI1 periodic saturation: a method for improving accuracy of quantitative perfusion imaging using pulsed arterial spin labeling. Magn Reson Med. Jun 1999;41(6): Kim SG. Quantification of relative cerebral blood flow change by flow-sensitive alternating inversion recovery (FAIR) technique: application to functional mapping. Magn Reson Med. Sep 1995;34(3): Tran TK, Vigneron DB, Sailasuta N, et al. Very selective suppression pulses for clinical MRSI studies of brain and prostate cancer. Magn Reson Med. Jan 2000;43(1): Ordidge RJ, Wylezinska M, Hugg JW, Butterworth E, Franconi F. Frequency offset corrected inversion (FOCI) pulses for use in localized spectroscopy. Magn Reson Med. Oct 1996;36(4): Andersen IK, Sidaros K, Gesmara H, Rostrup E, Larsson HB. A model system for perfusion quantification using FAIR. Magn Reson Imaging. Jun 2000;18(5): Calamante F, Thomas DL, Pell GS, Wiersma J, Turner R. Measuring cerebral blood flow using magnetic resonance imaging techniques. J Cereb Blood Flow Metab. Jul 1999;19(7):

142 CHAPTER VII CONCLUSION 128

143 SUMMARY Blood perfusion is vital to the survival of the brain since the brain has essentially no capacity for local energy storage and is dependent upon a constant and consistent blood supply to provide the constituents necessary to maintain function. Cerebral blood flow (CBF) measurements combined with physiological parameters can be used to reveal complex neurobiology in healthy and diseased individuals. Local perfusion change can be used to detect neural activity in cognitive and clinical neuroscience studies 1-7. A desirable technique for measuring CBF should consist of the following qualities: 1) quantitative, 2) non-invasive, 3) insensitive to motion errors, 4) high image resolution and quality, and 5) clinically acceptable imaging time. Over the past decades, numerous techniques have been proposed to measure CBF. From the development of radioactive isotopes in PET or SPECT imaging to the use of metal ion chelates in DSC-MRI or CT perfusion, significant improvements in terms of perfusion quantification, spatial resolution and invasiveness have been made. However, most perfusion techniques still rely on the use of an external contrast agent. The ultimate goal of a totally non-invasive technique that enables CBF mapping with high spatial and temporal resolution is still under development. During the development for advanced perfusion imaging techniques, a completely non-invasive MRI technique, known as arterial spin labeling (ASL), was developed for quantitative perfusion measurement. Better spatial resolution can be achieved with ASL compared to conventional external contrast-bolus based perfusion techniques (e.g. PET, DSC-MRI). The feasibility of ASL has been demonstrated for clinical neuroimaging applications (i.e. acute and chronic cerebrovascular disease, epilepsy, brain tumors, HIV, 129

144 neurodegenerative disorder and neuropsychiatric disease) and neuroscience studies (i.e. brain development, motor learning, aging, behavioral states and psychological stress) 1, 8. While in popular demand, ASL suffers from an inherently low SNR which limits its spatial and temporal resolution. In addition, a variety of parameters such as subject motion, system instability and magnetic susceptibility will significantly impact the quality of perfusion images and the accuracy of quantification. Those restrictions have limited ASL for routine clinical usage. The goal of this doctoral research work was to improve the ASL technique to overcome the aforementioned disadvantages, and thus to move ASL forward from the current status of a primary research tool to the clinical platform. ASL in the Clinical Environment The implementation of ASL in the clinical environment is expected to increase significantly in the near future, despite its current limitations. Here at Wake Forest University Baptist Medical Center, ASL perfusion images were acquired as a part of the routine clinical practice for neuroradiology. A fully automated pipeline 9 was developed to handle data transfer, image post-processing and display. Q2TIPS-FAIR 10, 11 with 2D EPI readout sequence was implemented as the ASL sequence. Since the initial installation of the pipeline in 2005, over 25,000 clinical cases have been acquired, which constitute the largest clinical ASL image database in the world to date. Those cases have revealed numerous pathological and physiological processes identified by the ASL technique that helped physician to make accurate diagnoses 2, 3, 7, 12, 13. The number of perfusion cases in the database reported here was counted in November,

145 From the early clinical experience, it was determined that variations in MR signal related to patient motion, system instability and disruption of the steady state can lead to dramatic artifacts in the perfusion images, rendering the final CBF maps unusable. A post-processing filter 14 was developed to detect and remove outliers in the perfusion weighted images to improve the quality of the final image (see Chapter II). The implementation of the filter has improved the diagnostic quality of the final images and the overall efficiency of the pipeline, including reducing motion or hardware induced artifacts and salvaging previously uninterpretable cases. The filter algorithm provided a fast and efficient solution to address the instability issue of ASL in a routine clinical environment with a large patient population. Current ASL Development The ASL technique consists of two components, the labeling (tagging) of the arterial blood and image acquisition, both of which are separate and independent processes. There exist various labeling schemes, each with its own advantages and shortcomings as described in Chapter I. Advancement in labeling schemes can improve perfusion sensitivity thus reducing the overall scan time. Improving the image acquisition can lead to higher image resolution and quality, as well as improved perfusion sensitivity. The following techniques were developed to improve the quality, sensitivity and stability of the ASL technique from both image acquisition and spin labeling. 131

146 3D GRASE PROPELLER (3DGP) Conventionally, fast imaging methods such as 2D EPI or spiral imaging are used as the ASL acquisition technique to capture the blood signal before its complete decay. There are several drawbacks associated with the 2D imaging, one of which is the limited brain coverage due to hemodynamic timing. Typically 2D images are acquired sequentially. The blood inflow time of each slice thus increases as the number of acquired slices increases, until the dynamic range of the inflow time is too large to ensure the labeled bolus is in the same vascular compartment in all slices. Some perfusion signal may also be lost due to cross-talk effects. Furthermore, 2D techniques such as EPI are highly sensitive to field inhomogeneity and magnetic susceptibility, which can lead to geometrical distortion and signal loss. Compared to 2D techniques, recent development of 3D acquisition techniques has evolved to overcome those limitations. Full brain coverage can be achieved easily with 3D techniques due to the simultaneous acquisition rather than separate sequential encodes. All slices of the brain are encoded together that provides an inherent advantage of higher SNR compared to 2D EPI 18. Furthermore, background suppression pulses 20 can be easily implemented with 3D techniques to reduce physiological noises and further improve perfusion image quality. Among various 3D techniques, 3D GRASE 18 has the advantage of refocusing dephased transverse signals for better signal-to-noise ratio (SNR) and higher image resolution. As a spin echo technique, it has the advantage of reduced susceptibility artifacts for improved image quality. However, one major drawback that limits the usage of 3D GRASE is the through-plane blurring caused by the T 2 decay as a result of the extended acquisition window. Furthermore, 3D GRASE is also susceptible to patient 132

147 motion, which according to our clinical experience, is one of the primary causes for degraded image quality. To address those problems, a novel image acquisition technique combining 3D GRASE and a PROPELLER trajectory (3DGP) was developed, described in Chapter III. Since signal averaging is necessary in ASL, the speculation is that rather than sampling k-space repetitively in the same pattern, it can be more beneficial to use an alternative trajectory for additional desirable properties. PROPELLER is chosen for several reasons. First, PROPELLER is well-known for its robustness against motion artifacts 21. Second, PROPELLER samples only the central rectangular region rather than the full k-space, thus having significantly reduced acquisition window duration compared to 3D GRASE. Consequently the through-plane blurring and off-resonance effects are greatly reduced per acquisition (shot). Since the final image resolution of PROPELLER is determined by the number of frequency encoding and the field of view, the image resolution can be improved with only minor increases in the acquisition window and the echo time. Those advantages of PROPELLER were fully demonstrated with the in-vivo experiments. The CBF maps acquired with 3DGP showed good agreements with 3D GRASE in terms of perfusion quantification and repeatability. The image quality of 3DGP is much greater than 3D GRASE, indicated by the following criteria: 1) perfusion signal in subcortical structures such as the thalamus and the caudate-putamen is more evident in 3DGP images; 2) off-resonance artifacts caused by nasal cavity in 3D GRASE are absent in 3DGP images; 3) through-plane blurring is significantly reduced exhibited by the detailed cortical structure in the coronal and sagittal view (see Chapter III, Figure 133

148 12). Overall, the experiment results support the hypothesis that alternative trajectory (i.e. PROPELLER) can be used to improve the image quality and the acquisition efficiency of single-shot 3D techniques (e.g. 3D GRASE) in ASL applications. Ghost Elimination via Spatial and Temporal Encoding (GESTE) One challenge in 3DGP reconstruction is the Nyquist ghost artifacts as a result of the EPI subsequence in the 3D GRASE readout. Nyquist ghosts, characterized as a faint copy of the imaging object shifted by one-half of the field of view, are caused by the misalignment between odd and even k-space lines (positive and negative readout gradients) due to the EPI zigzag trajectory. Conventional methods for correcting Nyquist ghosts rely on a reference scan to estimate and correct for the phase errors between odd and even readouts 24, 25. While acceptable for some applications, it is highly inefficient for 3DGP acquisition since a reference scan is required at every rotation angle. In addition, any phase variation during acquisition, even small ones, can cause significant residual ghost artifacts. A self-referenced Nyquist ghost correction method (GESTE), integrating two existing correction methods, PAGE 26 (spatial encoding) and PLACE 27 (temporal encoding), was developed and used for 3DGP post-processing, achieving better ghost suppression than previously reported methods 28. Implementation detail of GESTE is described in Chapter IV. The principle of GESTE acquisition is to use both spatial and temporal encoding for ghost removal. Spatial encoding is achieved with a multi-channel receiver coil, while temporal encoding is achieved by alternating the polarity of the readout gradients (G x ) on successive frames. The GESTE reconstruction consists of two steps: calibration and reconstruction. In the 134

149 calibration step, a ghost-free reference image is formed by interleaving and combining the two neighboring frames 27, 29, where GRAPPA 30 reconstruction coefficients for 2x accelerated data are then computed from the reference image. In the reconstruction step, the odd and even k-space lines of an individual frame, corresponding to either all positive or all negative readouts, are separated into two data sets, I odd and I even. This separation is equivalent to a 2x acceleration in each set. Two new images, I odd and I even, are then reconstructed by estimating the missing k-space lines in I odd and I even using GRAPPA and the reconstruction coefficients obtained from the calibration step. I odd and I even are then coherently combined after estimating and correcting for the zero and first order phase difference between I odd and I 31 even. Consequently, the final combined image is free of Nyquist ghosts and magnitude errors. It was shown from the results in Chapter IV that GESTE achieves superior ghost suppression compared to other correction methods. It is also robust against GRAPPA reconstruction errors and is able to maintain both the original SNR and temporal resolution. In the current implementation, GESTE acquires two temporally encoded data frames for every 3DGP brick. This does not lengthen the scan time in ASL imaging since the minimum signal average requirement to achieve sufficient perfusion SNR is more than double the Nyquist sampling requirement for the 3DGP acquisition (typical clinical PASL scans at 1.5T with 2D EPI acquire over 100 or more volumes 32 ). The total acquisition time remains the same since all data acquired contribute to the perfusion signal. This lies in contrast with reference scan based methods, where the reference data would not contribute to the final perfusion signal. 135

150 Pseudo-continuous ASL (PCASL) One major limitation associated with ASL is its low perfusion SNR. In humans, the arterial transit time for inflowing blood from the tagging plane to the imaging slices is comparable with T 1 of the blood 33. As a result, the signal of labeled spin is greatly reduced by the time the arterial blood reaches the brain tissue, resulting in low perfusion SNR and potentials for errors in perfusion quantification due to uncertain transit time. Similar problems exist in non-human primate (NHP) perfusion imaging. In addition, perfusion imaging in NHP requires higher spatial resolution in order to distinguish brain structures due to its smaller anatomical size, resulting in even lower SNR per voxel. Improvement in imaging acquisition alone is not sufficient enough to provide adequate perfusion SNR, advancement in spin labeling techniques is required. To overcome the limitations of conventional labeling techniques such as PASL and CASL (see Chapter I), an intermediate method, PCASL, was developed to combine the merits of PASL and CASL without inheriting the shortcomings. It has been shown that PCASL yields a 50% improvement in SNR over PASL due to closer tagging plane, a longer tagging bolus, and a higher tagging efficiency (~90%) compared to CASL 34 without the requirements for special hardware. Considering those advantages, balanced PCASL 34, 35 was implemented (see Chapter V) and tested in-vivo. The perfusion sensitivity was greatly improved in NHP experiments compared to PASL. The results have illustrated that it is feasible to use ASL technique (PCASL, with 3D readout sequence) to measure CBF in NHP. 136

151 Perfusion Phantom One difficulty in ASL technical development is the lack of a gold-standard. Nearly all new ASL techniques rely on previously published in-vivo results for method validation 36, 37. While widely accepted in the ASL development community, in-vivo validation is subject to many uncontrollable variables (e.g. physiological conditions) and thus is prone to biases and errors. The systematic biases among different ASL implementations may be constant, however, without a standard reference, it is difficult to find the accurate bias value and compare across subjects and methods quantitatively. A perfusion phantom that can mimic the cerebral perfusion process in a controlled environment (e.g. measurable flow rate, perfusion volume, etc) is the ideal candidate to fill the gap in the ASL developmental process. A flow-based, MR-compatible perfusion phantom was built by researchers at Virginia Polytechnic Institute and State University for validating ASL methods. The perfusion process is modeled using a cloth that separates the cylindrical body of the phantom into two compartments: a mixing chamber where the new fluid (inflowing arterial blood) can be fully mixed with the existing fluid (brain tissue), and a perfusion chamber that models the imaging state after the perfusion process has taken place. The phantom is validated using thermodynamics (see Chapter VI). The initial results obtained by thermodynamics have shown the phantom is capable of producing perfusion results within human physiological range. However, there are several design issues that are unaccounted for (see Chapter VI) which currently limit the MR experiments. Those issues are further discussed in the next section. 137

152 FUTURE DEVELOPMENT Things can always be improved upon. Any well-established techniques are the consequence of multiple iterations and improvements. Potential enhancements of aforementioned techniques are discussed in this section, with the purpose of aiding future development. Off-Resonance Correction in 3D GRASE PROPELLER (3DGP) Due to the shorter ETL per shot, there is less off-resonance error in a single 3DGP brick compared to a single 3D GRASE acquisition. However, the spatial and intensity distortions caused by the residual off-resonance effects still remain, and vary across all bricks due to the rotational trajectory. The combination of those bricks with incoherent off-resonance errors can cause destructive interference during image reconstruction. Consequently, the final measured perfusion SNR of 3DGP is lower than 3D GRASE, even though 3DGP has inherently higher signal energy based on pixel SNR calculation of the raw k-space data 38. This phenomenon is much worse at higher magnetic field, as the off-resonance effects increases with field strength. To recover the perfusion SNR for 3DGP, off-resonance correction must be incorporated into the reconstruction process. Off-resonance effect is a well-known phenomenon 39, and there have been numerous methods proposed to correct for off-resonance induced geometrical distortion 27, In particular, reverse gradient based methods 44, 45 are highly applicable to 3DGP acquisition. In short, reverse gradient methods are based on the fact that image distortion along the phase encoding direction is directly related to the polarity of the phase encoding gradients (i.e. the directions of k-space traversal). That is, the spatial distortion patterns 138

153 are reversed in two images acquired with phase gradients of opposite polarities. The undistorted images lie somewhere in between the two oppositely distorted images. Assuming the signal conservation along each phase encoding line, the undistorted image can then be estimated by proportionally shifting the signal from two distorted images via edge detection, line integration and higher order fitting 27, 42, 44, 45. Reverse gradient methods can be easily incorporated with 3DGP acquisition. A pair of PROPELLER blades 180 away from each other (e.g. 0 and 180, 90 and 270, 60 and 240 ) is in essence acquired with opposite phase encoding gradients. By rotating k-space data collection by a full 360, the reversed gradient method can be applied to correct for geometrical distortions in each pair of blades prior to PROPELLER reconstruction. Not only should this significantly improve the perfusion SNR for 3DGP, it should also improve the quality of image details as the in-plane blurring caused by combining incoherent distorted blades will also be corrected. Combining GESTE and ASL Encoding In the current implementation, GESTE and ASL encodings are two separate dimensions in the acquisition scheme. That is, for each PROPELLER angle, four bricks are acquired: a control brick and a label brick, each acquired twice with readout gradients of opposite polarities (for GESTE encoding). The minimum number of bricks needed per acquisition is thus N Min, Bricks = N PROPELLER * N ASL * N GESTE where N PROPELLER is the number of PROPELLER bricks, N ASL = 2 is the number of perfusion state (i.e. control and label), and N GESTE = 2 is number of GESTE encodings (i.e. positive and negative G x ). While the GESTE encodings do not pose a limitation on scan time due to large number of 139

154 data averaging needed for ASL, it may be possible to simultaneously encode both GESTE and ASL by coupling the readout gradient polarity change with the blood magnetization labeling state. The merge of the two encodings will reduce the total acquisition time by half. The additional time freed, which may still be needed for signal averaging, can now be used to encode other desirable properties. For instance, it can be used to encode additional bricks, thus allowing thinner bricks to be acquired to reduce off-resonance effects in each brick without violating the Nyquist sampling criterion; or it can be used to acquire extra data needed for off-resonance correction. The perfusion weighting in the GESTE encodings will not affect the Nyquist ghost removal process, since the tissue signal still dominates the image intensity, even in the presence of background suppression. Meanwhile, the toggling of the readout gradient will not affect ASL weighting since the GESTE reconstruction maintains the original temporal resolution. Hence, there is no restriction on how the two encoding schemes are combined. [Control with GESTE+, Label with GESTE-] or [Control with GESTE-, Label with GESTE+] will yield similar results. Overall, this is a more efficient way for data acquisition. Multi-Phase and Optimized PCASL Conventional ASL technique consists of two states in the acquisition scheme, control and label. For PCASL, the control and label states are defined by the phase difference between the two adjacent RF pulses in the labeling sequence. The phase of each tagging RF pulse can be written 46 as θ = θ + θ track offset 140

155 Figure 29. Simulated inversion response curve (in blue) as a function of the phase offsets according to 46. The control and label states are indicated by crosses and circles, respectively. The green color corresponds to the ideal situation when there is no phase error. The red color corresponds to a 45 phase error. It can be seen that the tagging efficiency is significantly reduced by the introduction of the phase error. where θ track accounts for the phase accumulated between the two successive RF pulses due to non-zero mean gradients and θ offset defines the control or label state. In the label state, θ offset is 0. In the control state, θ offset toggles between 0 and 180 in successive RF pulses during control state. As previously discussed in Chapter V, the PCASL labeling mechanism is highly sensitive to the accurate specification of phases between the inflowing spins and the tagging RF pluses. Factors that cause phase mismatch, such as imperfect gradient systems, eddy currents and off-resonance fields, can introduce an 141

156 Figure 30. MP-PCASL is illustrated with four-phases. The green crosses correspond to the ideal situation where no phase error is present. The red crosses correspond to a 45 phase error. In both cases, the correct perfusion signal can be estimated from both cases by fitting the measured signal to the inversion response curve in blue. Hence, MP-PCASL is robust against phase errors to improve the tagging efficiency. phase error term in θ which significantly compromises tagging efficiency, as shown in Figure 29. For an extremely large phase error (e.g. 90 ), the tagging efficiency can be reduced to zero. Multi-Phase PCASL (MP-PCASL), developed by Jung et. al 46, is proposed to reduce the sensitivity to these phase errors. Rather than acquiring the conventional two phases (i.e. θ offset = 0 and 180 ), multiple evenly distributed phase offsets are acquired (Figure 30). The data can then be fitted to a predefined inversion response curve 46 to estimate the true perfusion signal and the phase errors. The fitting procedure is done Tagging efficiency here is defined as the ratio of actual magnetization difference between control and label state and the ideal situation (i.e. difference between full relaxed and perfectly inverted magnetization). 142